architecture thesis on artificial intelligence

Grab your spot at the free arXiv Accessibility Forum

Help | Advanced Search

Computer Science > Machine Learning

Title: generative ai for architectural design: a literature review.

Abstract: Generative Artificial Intelligence (AI) has pioneered new methodological paradigms in architectural design, significantly expanding the innovative potential and efficiency of the design process. This paper explores the extensive applications of generative AI technologies in architectural design, a trend that has benefited from the rapid development of deep generative models. This article provides a comprehensive review of the basic principles of generative AI and large-scale models and highlights the applications in the generation of 2D images, videos, and 3D models. In addition, by reviewing the latest literature from 2020, this paper scrutinizes the impact of generative AI technologies at different stages of architectural design, from generating initial architectural 3D forms to producing final architectural imagery. The marked trend of research growth indicates an increasing inclination within the architectural design community towards embracing generative AI, thereby catalyzing a shared enthusiasm for research. These research cases and methodologies have not only proven to enhance efficiency and innovation significantly but have also posed challenges to the conventional boundaries of architectural creativity. Finally, we point out new directions for design innovation and articulate fresh trajectories for applying generative AI in the architectural domain. This article provides the first comprehensive literature review about generative AI for architectural design, and we believe this work can facilitate more research work on this significant topic in architecture.

Comments:	32 pages, 20 figures
Subjects:	Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
Cite as:	[cs.LG]
	(or [cs.LG] for this version)
	Focus to learn more arXiv-issued DOI via DataCite

Submission history

Access paper:.

HTML (experimental)
Other Formats

References & Citations

Google Scholar
Semantic Scholar

BibTeX formatted citation

Bibliographic and Citation Tools

Code, data and media associated with this article, recommenders and search tools.

Institution

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .

ArchiGAN: Artificial Intelligence x Architecture

First Online: 03 September 2020

Cite this chapter

architecture thesis on artificial intelligence

Stanislas Chaillou 6

6135 Accesses

42 Citations

AI will soon massively empower architects in their day-to-day practice. This article provides a proof of concept. The framework used here offers a springboard for discussion, inviting architects to start engaging with AI, and data scientists to consider Architecture as a field of investigation.

This is a preview of subscription content, log in via an institution to check access.

Access this chapter

Subscribe and save.

Get 10 units per month
Download Article/Chapter or eBook
1 Unit = 1 Article or 1 Chapter
Cancel anytime
Available as PDF
Read on any device
Instant download
Own it forever
Available as EPUB and PDF
Compact, lightweight edition
Dispatched in 3 to 5 business days
Free shipping worldwide - see info
Durable hardcover edition

Tax calculation will be finalised at checkout

Purchases are for personal use only

Institutional subscriptions

Big Architect, Little Architect

Introduction.

AI art in architecture

Zheng, H., & Huang, W. (2018). Architectural drawings recognition and generation through machine learning . Cambridge: MA, ACADIA.

Google Scholar

Peters, N. (2017). Master thesis: “Enabling alternative architectures: Collaborative frameworks for participatory design.” Cambridge, MA: Harvard Graduate School of Design.

Martinez, N. (2016). Suggestive drawing among human and artificial intelligences . Cambridge, MA: Harvard Graduate School of Design.

Download references

Author information

Authors and affiliations.

Spacemaker, Oslo, Norway

Stanislas Chaillou

You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Stanislas Chaillou .

Editor information

Editors and affiliations.

College of Architecture and Urban Planning, Tongji University, Shanghai, China

Philip F. Yuan

School of Engineering Cluster, Civil and Infrastructure Engineering, Melbourne, VIC, Australia

Rights and permissions

Reprints and permissions

Copyright information

About this chapter

Chaillou, S. (2020). ArchiGAN: Artificial Intelligence x Architecture. In: Yuan, P.F., Xie, M., Leach, N., Yao, J., Wang, X. (eds) Architectural Intelligence. Springer, Singapore. https://doi.org/10.1007/978-981-15-6568-7_8

Download citation

DOI : https://doi.org/10.1007/978-981-15-6568-7_8

Published : 03 September 2020

Publisher Name : Springer, Singapore

Print ISBN : 978-981-15-6567-0

Online ISBN : 978-981-15-6568-7

eBook Packages : Computer Science Computer Science (R0)

Share this chapter

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

Publish with us

Policies and ethics

Find a journal
Track your research

Computer Science
Artificial Intelligence

Use of Artificial Intelligence in the Field of Sustainable Architecture: Current Knowledge

Architecture Papers of the Faculty of Architecture and Design STU 26(1)
CC BY-NC-ND 3.0

Slovak University of Technology in Bratislava

Abstract and Figures

Discover the world's research

25+ million members
160+ million publication pages
2.3+ billion citations

Afif Fajar Zakariya
Aniekan Akpan Umoh
Chinedu Nnamdi Nwasike

Bindiya Jain
Shikha Sharma

César Frías
Jose María Peña
Érika Sánchez
Lorena Almeida

Yichun Huang

Xiangdong Zhou
AUTOMAT CONSTR

Recruit researchers
Join for free
Login Email Tip: Most researchers use their institutional email address as their ResearchGate login Password Forgot password? Keep me logged in Log in or Continue with Google Welcome back! Please log in. Email · Hint Tip: Most researchers use their institutional email address as their ResearchGate login Password Forgot password? Keep me logged in Log in or Continue with Google No account? Sign up

Hispanoamérica
Work at ArchDaily
Terms of Use
Privacy Policy
Cookie Policy

Pioneers: 6 Practices Bringing AI into Architecture

Written by Andreea Cutieru
Published on April 07, 2020

In this article, we tap into how AI could be augmenting, changing design processes, and how architects and other professionals are responding and incorporating these technological advancements into their design work. What kind of innovation can AI bring to this industry, and what has been experimented with so far? This selection of projects can help form an opinion on the architectural application of AI.

Artificial intelligence (AI) is a wide-ranging branch of computer science concerned with the development of systems able to perform tasks typically requiring human intelligence. The technology is advancing at a high pace, and it shows great potential for implementation across many fields. AI brings along opportunities that would radically change the existing workflow within the architecture profession. Although for now, the progress manifests itself at the fringes through research projects, art and cross-pollination between different study fields, it might not be long before this body of work reaches a critical mass, thus penetrating main-stream design workflow. Whether we are talking about well-established architecture practices, tech startups, architects with backgrounds in computer science, the following series illustrates the advent of AI in architecture.

AI+ Architecture / Stanislas Chaillou

Stanislas Chaillou's study of the potential of AI in space organization and architectural layouts has already gained significant notoriety. Through the use of deep learning and more specifically, GANs (Generative Adversarial Neural Networks), Chaillou developed a system that generates and furnishes floor plans, accounting for functionality and style. His thesis research is an example of viable tool architects might benefit from soon, as it would enable multiple iterations of the projects and produce legitimate floor plans that would, in turn, serve as the basis for further analysis and ideation by the designer. Such enhancement would allow for more informed decisions within a significantly reduced time frame.

Artificial Intelligence in Architecture / 3XN

Although this is solely a research project, the example shows how a renowned architecture practice positions itself with regards to emergent technology . In preparation for a paradigm shift, 3XN , through its research division GXN, has already undergone a thorough analysis of its design processes and AI technology. The research identified three major activities where AI could have a positive impact: Research – organizing information, Design- a better iterative process and Knowledge Management—developing an internal database of experience. The research project developed scenarios of what could be achieved within the next five years, with the primary objective of preparing the studio for the onset of AI into architecture practice so that the company can reflect on a proper response.

Daedalus Pavilion / Ai Build

AI Build, a London-based startup producing autonomous construction systems has teamed up with ARUP Engineers to create Daedalus Pavilion , a 5×5 metre latticework structure resembling a butterfly in flight. The project was meant to illustrate how robotics and artificial intelligence can change the future of the construction industry. The installation was built out of biodegradable filaments by Kuka construction robots. The latter used computer vision and machine learning algorithms to analyze any mistakes they made during construction and improvise solutions.

Plaza Life Revisited / XL Lab SWA Group

The firm's research division, XL Lab, revisits the findings of William H. Whyte's in his 1980 study The Social Life of Small Urban Spaces, aiming to understand what changed in the way people use the public realm today. The team employed a machine learning algorithm on video footage to develop heat maps describing dwell time, frequent and infrequent usage, and pedestrian counts. The Plaza Life Revisited research provides robust metrics to back up assumptions of how individuals occupy public space.

Generative Design Tool / Sidewalk Labs

Last year, Sidewalk Labs announced the development of a generative design tool that uses machine learning and computational design to create urban planning scenarios. Using geographical information, regulation, street layouts, orientation, weather patterns, building heights as input data, the tool generates a series of possible scenarios for architects and planners to assess and refine. With machine learning, the system has the ability to get better at the task and generate improved designs as it accumulates experience.

Ada / Jenny Sabin

Created as part of Microsoft’s Artist in Residence program, the installation Ada , named after first computer programmer Ada Lovelace, uses AI to create a performative environment. The first architectural pavilion project to incorporate AI, the exoskeleton translates data from visitor’s facial expressions and their voice tones into specific sentiments. The latter are then correlated with colours, spatial zones within the project, and responsive materials, to create an experience of photo-luminescence, exploring the idea of an interactive, genuinely responsive architecture.

The list of examples is by no means exhaustive, but through its diversity, it paints a comprehensive picture of the phenomenon. For the time being, the implementation of AI in architecture is driven mostly by tech companies and highly specialized firms, but architects are picking up the pace through research studies and installation projects. The penetration of Artificial Intelligence within the architectural field adds up to the profound changes that call for a reinvention of the profession.

Image gallery

Sustainability

想阅读文章的中文版本吗?

Daedalus Pavilion . Image Courtesy of Ai Build

科技建筑先驱者：6家引入人工智能的事务所

You've started following your first account, did you know.

You'll now receive updates based on what you follow! Personalize your stream and start following your favorite authors, offices and users.

LibGuides
A-Z List
Help

Artificial Intelligence

Background Information
Getting started
Browse Journals
Dissertations & Theses
Datasets and Repositories
Research Data Management 101
Scientific Writing
Find Videos
Related Topics
Quick Links
Ask Us/Contact Us

FIU dissertations

Non-FIU dissertations

Many universities provide full-text access to their dissertations via a digital repository. If you know the title of a particular dissertation or thesis, try doing a Google search.

Aims to be the best possible resource for finding open access graduate theses and dissertations published around the world with metadata from over 800 colleges, universities, and research institutions. Currently, indexes over 1 million theses and dissertations.

This is a discovery service for open access research theses awarded by European universities.

A union catalog of Canadian theses and dissertations, in both electronic and analog formats, is available through the search interface on this portal.

There are currently more than 90 countries and over 1200 institutions represented. CRL has catalog records for over 800,000 foreign doctoral dissertations.

An international collaborative resource, the NDLTD Union Catalog contains more than one million records of electronic theses and dissertations. Use BASE, the VTLS Visualizer or any of the geographically specific search engines noted lower on their webpage.

Indexes doctoral dissertations and masters' theses in all areas of academic research includes international coverage.

ProQuest Dissertations & Theses global

Related Sites

<< Previous: Browse Journals
Next: Datasets and Repositories >>
Last Updated: Jun 6, 2024 9:13 AM
URL: https://library.fiu.edu/artificial-intelligence

Information

Fiu libraries floorplans, green library, modesto a. maidique campus.

Floor		Resources
One
Two
Three
Four
Five
Six
Seven
Eight

Hubert Library, Biscayne Bay Campus

Floor		Resources
One
Two
Three

Directions: Green Library, MMC

Directions: Hubert Library, BBC

Areas of Study
Courses and Curriculum
Open Courses
Register for a Program
Certificate Program
Associate in Actuarial Science
Associate in Addiction Counseling
Associate in Agriculture Food And Resources
Associate in Animal Science
Associate in Anti Terrorism Security
Associate in Behavior Analysis In Special Education
Associate in Bioethics
Associate in Biosystems
Associate in Blockchain Technology & Digital Currency
Associate in Business Communication
Associate in Business Management
Associate in Chemistry
Associate in Climatology
Associate in Cloud Computing
Associate in Computer Engineering
Associate in Computer Programming
Associate in Computer Science
Associate in Criminal Justice
Associate in Culinary Arts
Associate in Cultural Theological Communication
Associate in Cybersecurity And Hacking
Associate in Data Communication And Networking
Associate in Database Administrator
Associate in Early Childhood Education
Associate in Ecotechnology
Associate in Ecotourism
Associate in Education
Associate in Educational Technology
Associate in Electric Vehicle Engineering
Associate in Electrochemical Engineering
Associate in Electronic Engineering
Associate in English Literature
Associate in Environmental Science
Associate in eVTOL Engineering
Associate in Fashion Design
Associate in Fine Arts
Associate in Foreign Trade
Associate in Genetic Engineering
Associate in Geography
Associate in Geology
Associate in Geophysical Sciences
Associate in Graphic Design
Associate in Health Sciences
Associate in History
Associate in Human Resources
Associate in Integrated Water Management
Associate in International Finance
Associate in International Marketing
Associate in Islamic Studies
Associate in Kinesiology And Physiotherapy
Associate in Library technology
Associate in Linguistics
Associate in Management
Associate in Management Information Systems
Associate in Maritime Management
Associate in Metallurgy
Associate in Micro and Multimode Grid Design
Associate in Modern Power and Energy Systems
Associate in Multimedia Design and Digital Art
Associate in Nutrition
Associate in Oil Gas And Energy Engineering
Associate in Operations Management
Associate in Optoelectronic Systems
Associate in Organizational Development
Associate in Organizational Diversity
Associate in Pedagogical Training For Professionals
Associate in Physical Culture And Sports
Associate in Physics
Associate in Public Health
Associate in Quantum Computing Technology
Associate in Radio And Television Production
Associate in Scenography
Associate in Social Media Marketing
Associate in Sociology
Associate in Sports Marketing
Associate in Sports Psychology
Associate in Sustainable Design and Construction
Associate in Sustainable Materials Science
Associate in Sustainable Natural Resources Management
Associate in Sustainable Tourism
Associate in Synthetic Biology
Associate in Virtual Archival Science
Associate of Adult Education
Associate of Advertising
Associate of Agriculture
Associate of American History
Associate of Biology
Associate of Chemical Engineering
Associate of Civil Engineering
Associate of Communications
Associate of Construction Management
Associate of Economics
Associate of Educational Administration
Associate of Electrical Engineering
Associate of Finance
Associate of Healthcare Administration
Associate of Human Resources Management
Associate of Humanities
Associate of Hydrology
Associate of Industrial Engineering
Associate of Information Systems
Associate of Information Technology
Associate of Interior Design
Associate of International Relations
Associate of Journalism
Associate of Legal Studies
Associate of Logistics
Associate of Marketing
Associate of Mass Media and Communication
Associate of Mathematics
Associate of Mechanical Engineering
Associate of Mining Engineering
Associate of Music
Associate of Nutrition Science
Associate of Philosophy
Associate of Physical Education
Associate of Political Science
Associate of Project Management
Associate of Psychology
Associate of Renewable Energy
Associate of Software Engineering
Associate of Sport Science
Associate of Statistics
Associate of Strategic Management
Associate of Technical Writing
Associate of Telecommunications
Associate of Theater
Associate of Theology
Associate of Tourism Planning and Development
Associate of Travel and Tourism
Associate of Unmanned Areal Systems Engineering
Associates in Artificial Intelligence
Associates in Engineering Systems
Associates in Physical Anthropology
Associates in Social Sciences
Associates in Sociocultural Anthropology
Associates in Systems Engineering
Associates of Accounting
Associates of Anthropology
Associates of Archeology
Associates of Architecture
Associates of Art History
Associates of Banking and Finance
Associates of Business Administration
Associates of Public Administration
Associates of Science
Associates of Urban Planning
Associates of Visual and Performing Arts
Micro and Multimode Grid Design
Bachelor in Data Communication and Networking
Bachelor in Actuarial Science
Bachelor in Addiction Counseling
Bachelor in Animal Science
Bachelor in Anti Terrorism Security
Bachelor in Artificial Intelligence
Bachelor in Arts in Cultural Theological Communication
Bachelor in Autonomous Vehicle Technology
Bachelor in Bachelor of Bioethics (BA)
Bachelor in Behavior Analysis in Special Education
Bachelor in Bibliotechnology
Bachelor in Biosystems
Bachelor in Blockchain Technology & Digital Currency
Bachelor in Business Communication
Bachelor in Business Management
Bachelor in Chemistry
Bachelor in Cloud Computing
Bachelor in Computer Science
Bachelor in Criminal Justice
Bachelor in Culinary Arts
Bachelor in Cybersecurity and Hacking
Bachelor in Database Administrator (BS)
Bachelor in Ecotechnology
Bachelor in Ecotourism
Bachelor in Education (B.Ed, BS)
Bachelor in Educational Technology
Bachelor in Electric Vehicle Engineering
Bachelor in Electrochemical Engineering
Bachelor in English Literature
Bachelor in eVTOL Engineering
Bachelor in Fashion Design (BA)
Bachelor in Fine Arts
Bachelor in Foreign Trade
Bachelor in Genetic Engineering
Bachelor in Geology
Bachelor in Geophysical Sciences
Bachelor in Graphic Design
Bachelor in Health Sciences
Bachelor in Integrated Water Management
Bachelor in International Finance
Bachelor in International Marketing
Bachelor in Islamic Studies
Bachelor in Kinesiology and Physiotherapy
Bachelor in Management
Bachelor in Metallurgy
Bachelor in Micro and Multimode Grid Design
Bachelor in Modern Power and Energy Systems
Bachelor in Multimedia Design and Digital Art
Bachelor in Nutrigenetics
Bachelor in Operations Management (BA)
Bachelor in Optoelectronic Systems
Bachelor in Organizational Development
Bachelor in Organizational Diversity
Bachelor in Physical Anthropology
Bachelor in Physical Culture And Sports
Bachelor in Public Relations
Bachelor in Quantum Computing Technology
Bachelor in Radio And Television Production
Bachelor in Scenography
Bachelor in Social Media Marketing
Bachelor in Social Sciences
Bachelor in Sociocultural Anthropology
Bachelor in Sport Management
Bachelor in Sports Marketing
Bachelor in Sports Psychology
Bachelor in Sustainable Design and Construction
Bachelor in Sustainable Materials Science
Bachelor in Sustainable Natural Resources Management
Bachelor in Sustainable Tourism
Bachelor in Synthetic Biology
Bachelor in Virtual Archival Science
Bachelor of Adult Education
Bachelor of Advertising
Bachelor of Animation
Bachelor of Art History
Bachelor of Biohacking and Nutrigenomics
Bachelor of Educational Administration
Bachelor of Healthcare Administration
Bachelor of Hydrology
Bachelor of Information Technology
Bachelor of Mass Media and Communication
Bachelor of Philosophy
Bachelor of Physical Education
Bachelor of Project Management
Bachelor of Public Administration
Bachelor of Software Engineering
Bachelor of Sport Science
Bachelor of Sports Science
Bachelor of Technical Writing
Bachelor of Theater
Bachelor of Theology
Bachelor of Tourism Planning and Development
Bachelor of Travel and Tourism
Bachelor of Unmanned Areal Systems Engineering
Bachelor of Urban Planning
Bachelor of Visual and Performing Arts
Bachelor of Web Design
Bachelors in Energy Storage and Battery Technology
Bachelors in Accounting
Bachelors in Accounting & Finance
Bachelors in Accounting and Auditing
Bachelors in Agriculture
Bachelors in Agronomy Engineering
Bachelors in Anthropology
Bachelors in Architecture
Bachelors in Automotive Engineering
Bachelors in Bacteriology
Bachelors in Banking and Finance
Bachelors in Biology
Bachelors in Business Administration
Bachelors in Chemical Engineering
Bachelors in Civil Engineering
Bachelors in Commercial Engineering
Bachelors in Communications
Bachelors in Computer Engineering
Bachelors in Construction Management
Bachelors in Criminal Justice
Bachelors in Criminology
Bachelors in Early Childhood Education
Bachelors in Economics
Bachelors in Electrical Engineering
Bachelors in Electromechanical Engineering
Bachelors in Electronic Engineering
Bachelors in Engineering
Bachelors in Environmental Science
Bachelors in Exterior Commerce
Bachelors in Finance
Bachelors in Finance and Banking
Bachelors in Financial Administration
Bachelors in Food Engineering
Bachelors in Global Supply Chain Management
Bachelors in Health Sciences
Bachelors in Healthcare Management
Bachelors in History
Bachelors in Hospitality Management
Bachelors in Human Development
Bachelors in Human Resource Management
Bachelors in Human Resources
Bachelors in Humanities
Bachelors in Industrial Engineering
Bachelors in Information Systems
Bachelors in Interior Design
Bachelors in International Business
Bachelors in International Relations
Bachelors in Journalism
Bachelors in Leadership Management
Bachelors in Legal Studies
Bachelors in Linguistics
Bachelors in Logistics
Bachelors in Marketing
Bachelors in Mathematics
Bachelors in Mechanical Engineering
Bachelors in Mine Engineering
Bachelors in Mining Engineering
Bachelors in Music
Bachelors in Nutrition
Bachelors in Nutritional Science
Bachelors in Oil Gas and Energy Engineering
Bachelors in Pedagogy
Bachelors in Physical Therapy
Bachelors in Physics
Bachelors in Political Science
Bachelors in Psychology
Bachelors in Public Health
Bachelors in Public Security
Bachelors in Renewable Energy
Bachelors in Social Communications
Bachelors in Sociology
Bachelors in Statistics
Bachelors in Strategic Management
Bachelors in Supply Chain Management
Bachelors in Systems Engineering
Bachelors in Telecommunications
Bachelors in Zoology
DBA Doctor of Business Administration
Doctor | in Actuarial Science
Doctor | in Agriculture Food And Resources
Doctor | in Animal Science
Doctor | in Biosystems
Doctor | in Cultural Theological Communication
Doctor | in Cybersecurity And Hacking
Doctor | in Early Childhood Education
Doctor | in Ecotechnology
Doctor | In Ecotourism
Doctor | In Educational Technology
Doctor | In Electronic Engineering
Doctor | In Foreign Trade
Doctor | of Biology (PhD)
Doctor in Addiction Counseling
Doctor in Behavior Analysis In Special Education
Doctor in Bibliotechnology
Doctor in Business Management
Doctor in Data Communication And Networking
Doctor of Adult Education
Doctor of Agriculture
Doctor of American History
Doctor of Animation
Doctor of Anthropology
Doctor of Archaeology
Doctor of Architecture (Ph.D.)
Doctor of Art History
Doctor of Artificial Intelligence
Doctor of Autonomous Vehicle Technology
Doctor of Biohacking and Nutrigenomics
Doctor of Business Management (DBM)
Doctor of Civil Engineering (D.Sc)
Doctor of Cloud Computing
Doctor of Economics (PhD)
Doctor of Educational Administration (PhD)
Doctor of Electric Vehicle Engineering
Doctor of Electrical Engineering (D.Sc, PhD)
Doctor of Electrochemical Engineering
Doctor of Engineering Systems (D.Sc)
Doctor of eVTOL Engineering
Doctor of Finance (PhD)
Doctor of Humanities
Doctor of Hydrology
Doctor of Industrial Engineering (D.Sc)
Doctor of International Relations (D.Sc)
Doctor of Legal Studies (PhD)
Doctor of Logistics
Doctor of Mass Media and Communication
Doctor of Mechanical Engineering (D.Sc)
Doctor of Micro and Multimode Grid Design
Doctor of Mining Engineering
Doctor of Music
Doctor of Nutrition Science
Doctor of Optoelectronic Systems
Doctor of Project Management (PhD)
Doctor of Public Administration
Doctor of Public Health (PhD, D.Sc)
Doctor of Quantum Computing
Doctor of Renewable Energy
Doctor of Sociology (PhD, D.Sc)
Doctor of Software Engineering
Doctor of Sport Management
Doctor of Sport Science
Doctor of Statistics
Doctor of Technical Writing
Doctor of Telecommunications (D.Sc)
Doctor of Theater
Doctor of Tourism Planning and Development
Doctor of Travel and Tourism
Doctor of Unmanned Aerial Systems Engineering
Doctor of Visual and Performing Arts
Doctor of Web Design
Doctoral Degree Programs
Doctoral in International Finance
Doctorate in Accounting
Doctorate in Actuarial Science
Doctorate in Adult Counseling
Doctorate in Advertising
Doctorate in Agriculture Food And Resources
Doctorate in Animal Science
Doctorate in Anti Terrorism Security
Doctorate in Behavior Analysis In Special Education
Doctorate in Bibliotechnology
Doctorate in Bioethics
Doctorate in Biosystems
Doctorate in Business Administration (DBA, PhD)
Doctorate in Business Communication
Doctorate in Business Management
Doctorate in Chemical Engineering (D.Sc)
Doctorate in Chemistry
Doctorate in Clinical Nutrition
Doctorate in Communication Online (D.Sc)
Doctorate in Computer Engineering
Doctorate in Computer Science
Doctorate in Conflict Resolution & Peace Building
Doctorate in Criminal Justice
Doctorate in Culinary Arts
Doctorate in Cultural Theological Communication
Doctorate in Cyber Security & Hacking
Doctorate in Data Communication And Networking
Doctorate in Digital Marketing
Doctorate in Early Childhood Education
Doctorate in Ecotechnology
Doctorate in Ecotourism
Doctorate in Education
Doctorate in Educational Technology
Doctorate in Electronic Engineering
Doctorate in Environmental Science
Doctorate in Fashion Design
Doctorate in Fine Arts
Doctorate in Foreign Trade
Doctorate in Genetic Engineering
Doctorate in Geophysical Sciences
Doctorate in Global Health
Doctorate in Health Sciences
Doctorate in Healthcare Administration Programs (PhD)
Doctorate in Hospital Administration
Doctorate in Human Resource Management (PhD)
Doctorate in Human Resources
Doctorate in Information Systems (D.Sc)
Doctorate in Information Technology (D.Sc)
Doctorate in Integrated Water Management
Doctorate in Interior Design (PhD)
Doctorate in International Marketing
Doctorate in International Relations
Doctorate in Islamic Studies
Doctorate in Journalism (PhD)
Doctorate in Kinesiology & Physiotherapy
Doctorate in Linguistics
Doctorate in Management
Doctorate in Maritime Management
Doctorate in Marketing (PhD)
Doctorate in Mathematics
Doctorate in Metallurgy
Doctorate in Modern Power and Energy Systems
Doctorate in Multimedia Design and Digital Art
Doctorate in Oil Gas And Energy Engineering
Doctorate in Organizational Development
Doctorate in Organizational Diversity
Doctorate in Pedagogical Training For Professionals
Doctorate in Physical Anthropology
Doctorate in Physical Culture And Sports
Doctorate in Physical Education (D.Sc)
Doctorate in Political Science
Doctorate in Psychology (PhD, DPsy)
Doctorate in Radio And Television Production
Doctorate in Scenography
Doctorate in Security Management
Doctorate in Social Media Marketing
Doctorate in Sociocultural Anthropology
Doctorate in Sports Management
Doctorate in Sports Marketing
Doctorate in Sports Psychology
Doctorate in Strategic Leadership
Doctorate in Strategic Management
Doctorate in Sustainable Design and Construction
Doctorate in Sustainable Materials Science
Doctorate in Sustainable Natural Resources Management
Doctorate in Sustainable Tourism
Doctorate in Synthetic Biology
Doctorate in Virtual Archival Science
Doctorate of Theology
DS – Doctorate in Science
Online Doctorate in Health Administration
Online Doctorate in Hospitality
Online Doctorate in Philosophy
Postdoctoral in Bioethics
School of Social and Human Studies
Master in Actuarial Science
Master in Addiction Counseling
Master in Agriculture Food And Resources
Master in Animal Science
Master in Anti Terrorism Security
Master in Autonomous Vehicle Technology
Master in Behavior Analysis In Special Education
Master in Bibliotechnology
Master in Bioethics
Master in Biosystems
Master in Blockchain Technology and Digital Currency
Master in Business Communication
Master in Climatology
Master in Cloud Computing
Master in Computer Programming
Master in Criminal Justice
Master in Culinary Arts
Master in Cultural Theological Communication
Master in Cybersecurity And Hacking
Master in Data Communication And Networking
Master in Database Administrator
Master in Early Childhood Education
Master in Ecotechnology
Master in Ecotourism
Master in Educational Technology
Master in Electric Vehicle Engineering
Master in Electronic Engineering
Master in Energy Storage and Battery Technology
Master in Engineering Systems (MS)
Master in eVTOL Engineering
Master in Fashion Design
Master in Fine Arts
Master in Foreign Trade
Master in Geography
Master in Geophysical Sciences
Master in Graphic Design
Master in History
Master in Integrated Water Management
Master in International Finance
Master in Islamic Studies
Master in Kinesiology And Physiotherapy
Master in Linguistics
Master in Management
Master in Management Information Systems
Master in Mass Media and Communication
Master in Metallurgy
Master in Micro and Multimode Grid Design
Master in Microbiology
Master in Modern Power and Energy Systems
Master in Multimedia Design and Digital Art
Master in Nutritional Science
Master in Oil Gas And Energy Engineering
Master in Organizational Development
Master in Organizational Diversity
Master in Pedagogical Training For Professionals
Master in Philosophy
Master in Physical Anthropology
Master in Physical Culture And Sports
Master in Quantum Computing
Master in Radio And Television Production
Master in Scenography
Master in Social Media Marketing
Master in Social Sciences
Master in Sociocultural Anthropology
Master in Software Engineering
Master in Sport Management
Master in Sport Science
Master in Sports Marketing
Master in Sports Psychology
Master in Statistics
Master in Strategic Management
Master in Sustainable Design and Construction
Master in Sustainable Materials Science
Master in Sustainable Natural Resources Management
Master in Sustainable Tourism
Master in Synthetic Biology
Master in Technical Writing
Master in Theater
Master in Theology
Master in Unmanned Aerial Systems Engineering (UAV and Drone Technology)
Master in Virtual Archival Science
Master of Adult Education
Master of Advertising (MS, MBA)
Master of Agriculture (MS)
Master of American History
Master of Animation
Master of Anthropology (MA)
Master of Archeology (MA)
Master of Artificial Intelligence
Master of Biohacking and Nutrigenomics
Master of Business Management (MS, MBM)
Master of Computer Engineering
Master of Construction Management
Master of Human Resources Management (MHRM)
Master of Humanities (MA)
Master of Hydrology (MS)
Master of International Relations (MS)
Master of Journalism (MA, MJ)
Master of Logistics (MA)
Master of Marketing (MS, MBA, MPA)
Master of Mining Engineering (MS)
Master of Music (MA)
Master of Physical Education (MS)
Master of Renewable Energy (MS)
Master of Science in Educational Administration (MS)
Master of Tourism Planning and Development
Master of Travel and Tourism
Master of Urban Planning
Master of Visual and Performing Arts
Master of Web Design
Masters in Accounting
Masters in Architecture
Masters in Art History
Masters in Banking and Finance
Masters in Biology
Masters in Business Administration
Masters in Business Management
Masters in Chemical Engineering
Masters in Chemistry
Masters in Civil Engineering
Masters in Clinical Psychology
Masters in Communication
Masters in Computer Science
Masters in Criminology
Masters in Economics
Masters in Education
Masters in Electrical Engineering
Masters in Environmental Science
Masters in Finance
Masters in Health Sciences
Masters in Healthcare Administration
Masters in Human Resources
Masters in Industrial Engineering
Masters in Information Systems
Masters in Information Technology
Masters in Interior Design
Masters in International Business
Masters in International Marketing
Masters in International Relations
Masters in Legal Studies
Masters in Mathematics
Masters in Mechanical Engineering
Masters in Nutrition
Masters in Political Science
Masters in Project Management
Masters in Psychology
Masters in Public Administration
Masters in Public Health
Masters in Sociology
Masters in Telecommunications
Postdoctoral in Actuarial Science
Postdoctoral in Addiction Counseling
Postdoctoral in Animal Science
Postdoctoral in Anti Terrorism Security
Postdoctoral in Autonomous Vehicle Technology
Postdoctoral in Behavior Analysis In Special Education
Postdoctoral in Bibliotechnology
Postdoctoral in Biohacking and Nutrigenomics
Postdoctoral in Biosystems
Postdoctoral in Blockchain Technology and Digital Currency
Postdoctoral in Business Communication
Postdoctoral in Business Management
Postdoctoral in Chemistry
Postdoctoral in Cloud Computing
Postdoctoral in Computer Engineering
Postdoctoral in Computer Science
Postdoctoral in Criminal Justice
Postdoctoral in Culinary Arts
Postdoctoral in Cultural Theological Communication
Postdoctoral in Cybersecurity And Hacking
Postdoctoral in Data Communication And Networking
Postdoctoral in Early Childhood Education
Postdoctoral in Ecotechnology
Postdoctoral in Ecotourism
Postdoctoral in Education
Postdoctoral in Educational Technology
Postdoctoral in Electrical Vehicle Engineering
Postdoctoral in Electrochemical Engineering
Postdoctoral in Electronic Engineering
Postdoctoral in Energy Storage and Battery Technology
Postdoctoral in Engineering Systems
Postdoctoral in English Literature
Postdoctoral in Environmental Science
Postdoctoral in eVTOL Engineering
Postdoctoral in Fashion Design
Postdoctoral in Fine Arts
Postdoctoral in Foreign Trade
Postdoctoral in Genetic Engineering
Postdoctoral in Geophysical Sciences
Postdoctoral in Graphic Design
Postdoctoral in Health Sciences
Postdoctoral in Human Resources
Postdoctoral in Humanities
Postdoctoral in Integrated Water Management
Postdoctoral in International Finance
Postdoctoral in International Marketing
Postdoctoral in International Relations
Postdoctoral in Islamic Studies
Postdoctoral in Journalism
Postdoctoral in Kinesiology And Physiotherapy
Postdoctoral in Linguistics
Postdoctoral in Logistics
Postdoctoral in Management
Postdoctoral in Marketing
Postdoctoral in Mass Communication
Postdoctoral in Mathematics
Postdoctoral in Metallurgy
Postdoctoral in Micro and Multimode Grid Design
Postdoctoral in Mining Engineering
Postdoctoral in Modern Power and Energy Systems
Postdoctoral in Multimedia Design and Digital Art
Postdoctoral in Music
Postdoctoral in Nutrigenetics
Postdoctoral in Nutrition
Postdoctoral in Oil, Gas and Energy Engineering
Postdoctoral in Organizational Development
Postdoctoral in Organizational Diversity
Postdoctoral in Pedagogical Training For Professionals
Postdoctoral in Philosophy
Postdoctoral in Physical Anthropology
Postdoctoral in Physical Culture And Sports
Postdoctoral in Political Science
Postdoctoral in Public Administration
Postdoctoral in Public Health
Postdoctoral in Quantum Computing
Postdoctoral in Radio And Television Production
Postdoctoral in Social Media Marketing
Postdoctoral in Social Sciences
Postdoctoral in Sociocultural Anthropology
Postdoctoral in Sociology
Postdoctoral in Software Engineering
Postdoctoral in Sport Management
Postdoctoral in Sport Science
Postdoctoral in Sports Marketing
Postdoctoral in Sports Psychology
Postdoctoral in Statistics
Postdoctoral in Strategic Management
Postdoctoral in Sustainable Design and Construction
Postdoctoral in Sustainable Materials Science
Postdoctoral in Sustainable Natural Resources Management
Postdoctoral in Sustainable Tourism
Postdoctoral in Synthetic Biology
Postdoctoral in Technical Writing
Postdoctoral in Theater
Postdoctoral in Theology
Postdoctoral in Virtual Archival Science
Postdoctoral of Biology
Postdoctoral of Industrial Engineering
Postdoctoral of Information Systems
Postdoctoral of Legal Studies
Postdoctoral Research in Accounting
Postdoctoral Research in Adult Education
Postdoctoral Research in Advertising
Postdoctoral Research in Agriculture
Postdoctoral Research in American History
Postdoctoral Research in Animation
Postdoctoral Research in Anthropology
Postdoctoral Research in Archeology
Postdoctoral Research in Architecture
Postdoctoral Research in Art History
Postdoctoral Research in Artificial Intelligence
Postdoctoral Research in Banking and Finance
Postdoctoral Research in Business Administration
Postdoctoral Research in Business Management
Postdoctoral Research in Chemical Engineering
Postdoctoral Research in Civil Engineering
Postdoctoral Research in Communications
Postdoctoral Research in Economics
Postdoctoral Research in Educational Administration
Postdoctoral Research in Electrical Engineering
Postdoctoral Research in Finance
Postdoctoral Research in Healthcare Administration
Postdoctoral Research in Human Resources Management
Postdoctoral Research in Information Technology
Postdoctoral Research in Interior Design
Postdoctoral Research in International Relations
Postdoctoral Research in Mechanical Engineering
Postdoctoral Research in Physical Education
Postdoctoral Research in Project Management
Postdoctoral Research in Psychology
Postdoctoral Research in Renewable Energy
Postdoctoral Research in Scenography
Postdoctoral Research in Telecommunications
Postdoctoral Research in Tourism Planning and Development
Postdoctoral Research in Travel and Tourism
Postdoctoral Research in Unmanned Aerial Systems Engineering (UAV and drone technology)
Postdoctoral Research in Urban Planning
Postdoctoral Research in Visual and Performing Arts
Postdoctoral Research in Web Design
Postdoctoral Research Program

Distance Learning at AIU is enhanced by vast academic resources and innovative technologies build into the Virtual Campus: Hundreds of self-paced courses with video lectures and step by step lessons, thousands of optional assignments, 140,000 e-books, the Social Media & Networking platform allowing collaboration/chat/communications between students, and MYAIU develop students holistically in 11 areas beyond just academics.

The world is YOUR campus!”, that is the message of AIU’s month magazine Campus Mundi. Hear the voices and see the faces that make up AIU. Campus Mundi brings the world of AIU to you every months with inspirational stories, news and achievements by AIU members from around the world (students and staff are located in over 200 countries).

Please enter your credentials

Student Advisor Administrator Admissions Quiz Editor Link Exchange

keep me logged-in

The Transformative Power of Artificial Intelligence: A Deep Dive into AI

Artificial Intelligence (AI) has rapidly evolved from a niche area of computer science to a cornerstone of modern technology. Its applications span numerous fields, including healthcare, finance, transportation, and entertainment. This article explores the fundamental concepts of AI, its diverse applications, and the ethical considerations surrounding its development and deployment.

Understanding Artificial Intelligence

Artificial Intelligence refers to the simulation of human intelligence in machines that are programmed to think and learn. These systems can perform tasks that typically require human intelligence, such as recognizing speech, making decisions, and translating languages. AI is broadly classified into two categories:

Narrow AI: Also known as Weak AI, it is designed to perform a narrow task (e.g., facial recognition, internet searches, or self-driving cars). It operates under a limited pre-defined set of constraints and parameters.
General AI: Also known as Strong AI, it possesses the ability to understand, learn, and apply intelligence across a wide range of tasks. This type of AI remains theoretical and is the ultimate goal of AI research.

Core Components of AI

AI integrates various technologies and methodologies to function effectively. Key components include:

Machine Learning (ML): A subset of AI focused on developing algorithms that allow computers to learn from and make predictions based on data.
Neural Networks: These are computational models inspired by the human brain, consisting of interconnected nodes (neurons) that process information in layers.
Natural Language Processing (NLP): A branch of AI that deals with the interaction between computers and humans through natural language.
Computer Vision: This field enables computers to interpret and make decisions based on visual data from the world.

Applications of AI

AI’s versatility is showcased in its wide-ranging applications, transforming industries and enhancing human capabilities.

1. Healthcare

AI is revolutionizing healthcare by enabling early disease detection, personalized treatment plans, and efficient management of health records. AI algorithms can analyze medical images to identify diseases like cancer at an early stage, while predictive analytics can anticipate patient needs and improve outcomes.

In the finance sector, AI is used for fraud detection, algorithmic trading, risk management, and customer service. AI systems can analyze vast amounts of data to identify fraudulent transactions and predict market trends, providing significant advantages to financial institutions.

3. Transportation

AI powers autonomous vehicles, improving safety and efficiency in transportation. Self-driving cars use AI to process data from sensors and cameras to navigate roads and avoid obstacles. AI also optimizes logistics and supply chain management, reducing costs and delivery times.

4. Entertainment

In the entertainment industry, AI enhances user experiences through personalized recommendations on platforms like Netflix and Spotify. AI-driven content creation tools are also emerging, helping artists and creators produce music, art, and literature.

Ethical Considerations in AI

As AI technology advances, it raises important ethical questions and challenges that need to be addressed:

Bias and Fairness: AI systems can perpetuate and even amplify biases present in training data, leading to unfair treatment of certain groups. Ensuring fairness and transparency in AI algorithms is crucial.
Privacy: AI’s ability to process and analyze large amounts of data raises concerns about data privacy and surveillance. Balancing innovation with the protection of individual privacy is essential.
Job Displacement: The automation of tasks through AI can lead to job displacement, affecting livelihoods. It is important to consider strategies for workforce transition and reskilling.
Accountability: Determining accountability for decisions made by AI systems, especially in critical areas like healthcare and criminal justice, is a significant challenge. Establishing clear guidelines and regulations is necessary.

The Future of AI

The future of AI holds immense potential and promise. Ongoing research aims to advance AI capabilities, making systems more robust, intelligent, and ethical. Collaboration between technologists, policymakers, and ethicists will be key to navigating the challenges and maximizing the benefits of AI.

Artificial Intelligence is transforming the way we live and work, driving innovation across multiple sectors. As AI continues to evolve, it offers exciting opportunities to improve our lives while posing significant ethical challenges. By prioritizing responsible development and deployment, we can harness the power of AI to create a better, more equitable future for all.

The Importance of Employee Well-Being in Today’s Workplace

The Role and Importance of Information Technology (I.T.) in Today’s World

The Rise of ChatGPT and AI: Transforming Communication and Beyond

The Transformative Impact of Artificial Intelligence: Applications and Benefits

The Impact of AI on Modern Society

ChatGPT: Revolutionizing Human-Computer Interaction

Popular Tags

Contact Us Atlantic International University

Quick links.

Artificial Intelligence

Research Groups/Events

Related News

The study of systems that behave intelligently, artificial intelligence includes several key areas where our faculty are recognized leaders: computer vision, machine listening, natural language processing, machine learning and robotics.

Computer vision systems can understand images and video, for example, building extensive geometric and physical models of cities from video, or warning construction workers about nearby dangers. Natural language processing systems understand written and spoken language; possibilities include automatic translation of text from one language to another, or understanding text on Wikipedia to produce knowledge about the world. Machine listening systems understand audio signals, with applications like speech recognition, acoustic monitoring, or transcribing polyphonic music automatically. Crucial to modern artificial intelligence, machine learning methods exploit examples in order to adjust systems to work as effectively as possible. Robotics puts artificial intelligence into practice using machines that perceive and interact with the physical world.

Strengths and Impact

The AI group at Illinois is strong, diverse, and growing. It combines expertise in core strengths with promising new research directions.

Research Focus

In machine learning, AI group faculty are studying theoretical foundations of deep and reinforcement learning; developing novel models and algorithms for deep neural networks, federated and distributed learning; as well as investigating issues related to scalability, security, privacy, and fairness of learning systems. Computer vision faculty are developing novel approaches for 2D and 3D scene understanding from still images and video; joint understanding of images and language; low-shot learning (recognition of rare or previously unseen categories); transfer learning and domain adaptation (adapting pre-trained systems to a changing data distribution); and image generation and editing approaches based on generative neural networks. Natural language processing faculty are working on topics such as grounded language understanding, information extraction and text mining, and knowledge-driven natural language generation for applications such as scientific discovery. Machine listening faculty are working on sound and speech understanding, source separation, and enhancement, as well as applications in music and computing. Robotics faculty are developing novel planning algorithms for grasping, locomotion, and navigation; investigating multi-robot systems; as well as pursuing high-impact applications of robotics to medicine, agriculture, home care, and autonomous driving.

Research Awards

The excellence and impact of the AI group’s research has been recognized by a number of awards, including NSF CAREER (Amato, Hauser, Hockenmaier, Hoiem, Ji, Koyejo, Lazebnik, Smaragdis, Telgarsky), Sloan Research Fellowship (Hoiem, Koyejo, Lazebnik), Microsoft Research Faculty Fellowship (Lazebnik), AFOSR Young Investigator (Chowdhary), IEEE PAMI Significant Young Researcher Award (Hoiem), MIT TR-35 (Li, Smaragdis), Intel Rising Star Award (Li), “Young Scientist” selected by World Economic Forum (Ji), “AI’s Top 10 to Watch” Award by IEEE Intelligent Systems (Ji), ACM Fellow (Amato, Forsyth, Warnow), IEEE Fellow (Amato, Forsyth, Lazebnik, Smaragdis), IEEE Technical Achievement Award (Forsyth), and Packard Fellowship (Warnow).

In the last few years, AI group members received a number of best paper awards, including: IEEE Signal Processing Society Best Paper Award (Smaragdis, 2018 and 2020), IEEE MLSP Best Paper Award (Smaragdis, 2017), Best Demo Paper Award at the 58th Annual Meeting of the Association for Computational Linguistics (Ji, 2020).

Group Research

AI group research has led to a number of startups. Derek Hoiem is co-founder and Chief Science Officer of Reconstruct, which visually documents construction sites, matching images to plans and analyzing productivity and risk for delay. Girish Chowdhary is co-founder and CTO of EarthSense, a startup creating machine learning and robotics solutions for agriculture, whose work was featured in a 2020 New York Times article. David Forsyth advises a number of startups focusing on augmented reality and image synthesis, including Lightform, Revery, and Depix.

AI faculty are playing key roles in two $20 million AI institutes recently funded by the National Science Foundation and the U.S. Department of Agriculture’s National Institute of Food and Agriculture. The AI Institute for Future Agricultural Resilience, Management, and Sustainability (AIFARMS), led by Vikram Adve from CS, features Romit Chowdhary as Associate Director of Research, with other investigators including Alexander Schwing, Katherine Driggs-Campbell, Indranil Gupta, Kris Hauser, Julia Hockenmaier, Heng Ji, Sanmi Koyejo, and Paris Smaragdis. The AI Institute for Molecular Discovery, Synthetic Strategy, and Manufacturing, led by Huimin Zhao from Chemical Engineering, involves Heng Ji and Jian Peng as investigators.

Research Efforts and Groups

Beckman Institute
Center for Artificial Intelligence Innovation (NCSA)
Deep Learning Major Research Instrument Project (NCSA)
Natural Language Processing Group
Speech and Language Engineering Group
Center for Autonomy
Robotics Group
Robotics Seminar Series (Friday) and student mailing list
NLP: reading group, seminar
Computer Vision: mailing list, vision lunch (Thursday), external speaker series (Tuesday)
Illinois Computer Science Speaker Series : brings prominent leaders and experts to campus to share their ideas and promote conversations about important challenges and topics in the discipline.

Faculty & Affiliate Faculty

Robot Motion and Task Planning, Multi-Agent Systems, Crowd Simulation

Machine Learning Methods for Imaging Science, Image Reconstruction, Deep Learning for Inverse Problems

Human-Centered Natural Language Processing, AI for Science, Adaptive Language Interfaces

Machine Learning, Learning Theory, Optimization, Generative Models, Sequential Decision Making, Physics-Guided Machine Learning, Differential Privacy

Motion Planning and Control

Machine Learning, Natural Language Processing, AI Applications, Data Management Support for AI

Control, Autonomy and Decision Making, Vision and LIDAR Based Perception, GPS Denied Navigation

Computational Statistics, Reproducible Research, Statistics Education, Machine Learning

Social Network Analysis, Natural Language Processing, Machine Learning

Signal Processing, Computational Imaging, Machine Perception, Data Science

Autonomous Vehicles, Validating Autonomous Systems, Interactive Control Policies for Intelligent Systems in Multi-Agent Settings

Computational Linguistics

Computer Vision, Object Recognition, Scene Understanding

Computer Vision Analytics for Building and Construction Performance Monitoring

Computer Vision, Machine Learning, Motion Analysis, Robotics

Computer Vision, Robotics, Machine Learning

Conversational AI and Natural Language Processing

Machine Learning, Natural Language-Based Text Analysis, Text Summarization

Statistical Speech Technology

Motion Planning, Optimal Control, Integrated Planning and Learning, Robot Systems

Natural Language Processing, Computational Linguistics

Computer Vision, Object Recognition, Spatial Understanding, Scene Interpretation

Probabilistic Graphical Models; Deep Learning; Data Science; Health Analytics; Safety, Reliablity and Security of Autonomous Systems; Reinforcement Learning

ML4Code, ML interpretability, testing, and debugging

Natural Language Processing, especially on Information Extraction and Knowledge-driven Natural Language Generation, Text Mining, Knowledge Graph Construction for Scientific Discovery

Reinforcement Learning, Machine Learning, Sample Complexity Analyses

Analytics with Machine Learning, Databases with Machine Learning, Machine Learning Security, Machine Learning + Cryptography

HCI for ML, AI Explainability

Cyberinfrastructure for Machine Learning, Machine Learning Systems Research, Deep Learning Applications

Systems for Machine Learning, Machine Learning for Systems

Computer Vision, Scene Understanding, Visual Learning, Vision and Language

Adversarial Machine Learning, Robust Learning

Cyberinfrastructure for Digital Preservation, Auto-Curation, and Managing Unstructured Digital Collections

AI for Audio; Model Compression; Personalized AI; Signal Separation, Enhancement, and Coding

Motion Planning and Control, Autonomous Robots

Natural Language Processing, Machine Learning, Large Language Models, AI for Science

Machine Learning and Optimization

Computer Vision, Ego4D, VR/AR, Mobile Health, Health AI, Machine Learning, Developmental Machine Learning, Behavioral Imaging

Field Robotics, Autonomous Systems Engineering, Machine Perception, Computer Vision

Machine Translation, Computational Morphology & Syntax

Machine Learning, Computer Vision

Certified Artificial Intelligence, Adversarial Robustness, Neural Network Verification, Safe Deep Learning

Machine Learning for Audio, Speech and Music; Signal Processing; Source Separation; Sound Recognition and Classification

Deep Learning for Drug Discovery, Clinical Trial Optimization, Computational Phenotyping, Clinical Predictive Modeling, Mobile Health and Health Monitoring, Tensor Factorization, and Graph Mining

Deep Learning Theory

Explainable AI, Fairness in AI, Adversarial Maching Learning

Conversational AI

Computer Vision, Robotics

Computer Vision, Machine Learning, Meta-Learning, Robotics

Machine Learning in Computational Genomics, Ensemble Methods, Statistical Estimation

Efficient DL/AI Systems and Algorithms, Parallel Computing and Runtime, Natural Language Processing, AI for Science

Machine Learning Theory and Applications, Optimization, Reinforcement Learning, Robustness, Generative AI, Large Language Models

Machine Learning, Representation Learning, Algorithmic Fairness, Probabilistic Models

Adjunct Faculty

Machine Learning, Automatic Reasoning

Machine Learning, Neuroimaging, Biomedical Imaging

Machine Learning, Natural Language Processing, Knowledge Representation, Reasoning

CS professor Wang selected to participate in NAE 2024 Frontiers of Engineering Symposium

July 22, 2024

An Illinois CS team is giving robots a sense of touch

July 18, 2024

CS professors Godfrey and Wang join project building an intelligent and self-adaptive operating system

May 24, 2024

CS professor Tianyin Xu and team develop a "push button" for testing cloud systems

May 21, 2024

Load more news

Who’s Teaching What
Subject Updates
MEng program
Opportunities
Minor in Computer Science
Resources for Current Students
Program objectives and accreditation
Graduate program requirements
Admission process
Degree programs
Graduate research
EECS Graduate Funding
Resources for current students
Student profiles
Instructors
DEI data and documents
Recruitment and outreach
Community and resources
Get involved / self-education
Rising Stars in EECS
Graduate Application Assistance Program (GAAP)
MIT Summer Research Program (MSRP)
Sloan-MIT University Center for Exemplary Mentoring (UCEM)
Electrical Engineering
Computer Science
Artificial Intelligence + Decision-making
AI and Society
AI for Healthcare and Life Sciences
Artificial Intelligence and Machine Learning
Biological and Medical Devices and Systems
Communications Systems
Computational Biology
Computational Fabrication and Manufacturing
Computer Architecture
Educational Technology
Electronic, Magnetic, Optical and Quantum Materials and Devices
Graphics and Vision
Human-Computer Interaction
Information Science and Systems
Integrated Circuits and Systems
Nanoscale Materials, Devices, and Systems
Natural Language and Speech Processing
Optics + Photonics
Optimization and Game Theory
Programming Languages and Software Engineering
Quantum Computing, Communication, and Sensing
Security and Cryptography
Signal Processing
Systems and Networking
Systems Theory, Control, and Autonomy
Theory of Computation
Departmental History
Departmental Organization
Visiting Committee
News & Events
News & Events
EECS Celebrates Awards

Doctoral Thesis: Parameterizations of Neural Fields

32-G449 (Patil/Kiva)

By: Clinton Wang

Supervisor: Polina Golland

Date: Wednesday, August 7
Time: 10:00 am - 12:00 pm
Category: Thesis Defense
Location: 32-G449 (Patil/Kiva)

Additional Location Details:

Abstract: Neural fields are an invaluable tool in the modern repertoire of signal representations, finding success in diverse applications across many types of signals. The first part of this talk explores the design space of neural field parameterizations, focusing on two very different tasks—novel view synthesis in large 3D scenes, and motion stabilization in volumetric time series. We describe techniques to make neural fields better matched to these signals and task requirements. Then, we describe methods for using neural fields as datapoints for data-driven learning, which addresses a key shortcoming of neural fields relative to conventional signal representations. Since the heterogeneity of neural field designs makes most data-driven approaches unusable, we introduce a sampling-based approach that is agnostic to how the field is parameterized. This method enables tasks like classification or representation learning to be performed on neural fields analogously to discrete signals like images.

REVIEW article

From outputs to insights: a survey of rationalization approaches for explainable text classification.

$\r\nErick Mendez Guzman$

1 Department of Computer Science, The University of Manchester, Manchester, United Kingdom
2 ASUS Intelligent Cloud Services (AICS), ASUS, Singapore, Singapore

Deep learning models have achieved state-of-the-art performance for text classification in the last two decades. However, this has come at the expense of models becoming less understandable, limiting their application scope in high-stakes domains. The increased interest in explainability has resulted in many proposed forms of explanation. Nevertheless, recent studies have shown that rationales , or language explanations, are more intuitive and human-understandable, especially for non-technical stakeholders. This survey provides an overview of the progress the community has achieved thus far in rationalization approaches for text classification. We first describe and compare techniques for producing extractive and abstractive rationales. Next, we present various rationale-annotated data sets that facilitate the training and evaluation of rationalization models. Then, we detail proxy-based and human-grounded metrics to evaluate machine-generated rationales. Finally, we outline current challenges and encourage directions for future work.

1 Introduction

Text classification is one of the fundamental tasks in Natural Language Processing (NLP) with broad applications such as sentiment analysis and topic labeling, among many others ( Aggarwal and Zhai, 2012 ; Vijayan et al., 2017 ). Over the past two decades, researchers have leveraged the power of deep neural networks to improve model accuracy for text classification ( Kowsari et al., 2019 ; Otter et al., 2020 ). Nonetheless, the performance improvement has come at the cost of models becoming less understandable for developers, end-users, and other relevant stakeholders ( Danilevsky et al., 2020 ). The opaqueness of these models has become a significant obstacle to their development and deployment in high-stake sectors such as the medical ( Tjoa and Guan, 2020 ), legal ( Bibal et al., 2021 ), and humanitarian domains ( Mendez et al., 2022 ).

As a result, Explainable Artificial Intelligence (XAI) has emerged as a relevant research field aiming to develop methods and techniques that allow stakeholders to understand the inner workings and outcome of deep learning-based systems ( Gunning et al., 2019 ; Arrieta et al., 2020 ). Several lines of evidence suggest that providing insights into text classifiers' inner workings might help to foster trust and confidence in these systems, detect potential biases or facilitate their debugging ( Arrieta et al., 2020 ; Belle and Papantonis, 2021 ; Jacovi and Goldberg, 2021 ).

One of the most well-known methods for explaining the outcome of a text classifier is to build reliable associations between the input text and output labels and determine how much each element (e.g., word or token) contributes toward the final prediction ( Hartmann and Sonntag, 2022 ; Atanasova et al., 2024 ). Under this approach, methods can be divided into feature importance score-based explanations ( Simonyan et al., 2014 ; Sundararajan et al., 2017 ), perturbation-based explanations ( Zeiler and Fergus, 2014 ; Chen et al., 2020 ), explanations by simplification ( Ribeiro et al., 2016b ) or language explanations ( Lei et al., 2016 ; Liu et al., 2019a ). It is important to note that the categories cited above are not mutually exclusive, and explainability methods can combine several. This is exemplified in the work undertaken by Ribeiro et al. (2016a) , who developed the Local Interpretable Model-Agnostic Explanations method (LIME) combining perturbation-based and explanations by simplification.

Rationalization methods attempt to explain the outcome of a model by providing a natural language explanation ( rationale ; Lei et al., 2016 ). It has previously been observed that rationales are more straightforward to understand and easier to use since they are verbalized in human-comprehensible natural language ( DeYoung et al., 2020 ; Wang and Dou, 2022 ). It has been shown that for text classification, annotators look for language cues within a text to support their labeling decisions at a class level ( human rationales ; Chang et al., 2019 ; Strout et al., 2019 ; Jain et al., 2020 ).

Rationales for explainable text classification can be categorized into extractive and abstractive rationales ( Figure 1 ). On the one hand, extractive rationales are a subset of the input text that support a model's prediction ( Lei et al., 2016 ; DeYoung et al., 2020 ). On the other hand, abstractive rationales are texts in natural language that are not constrained to be grounded in the input text. Like extractive rationales, they contain information about why an instance is assigned a specific label ( Camburu et al., 2018 ; Liu et al., 2019a ).

Figure 1 . Example of an extractive and abstractive rationale supporting the sentiment classification for a movie review.

This survey refers to approaches where human rationales are not provided during training, as unsupervised rationalization methods ( Lei et al., 2016 ; Yu et al., 2019 ). In contrast, we refer to those for producing rationales where human rationales are available as additional supervision signal during training, as supervised rationalization methods ( Bao et al., 2018 ; DeYoung et al., 2020 ; Arous et al., 2021 ).

Even though XAI is a relatively new research field, several studies have begun to survey explainability methods for NLP. Drawing on an extensive range of sources, Danilevsky et al. (2020) and Zini and Awad (2022) provided a comprehensive review of terminology and fundamental concepts relevant to XAI for different NLP tasks without going into the technical details of any existing method or taking into account peculiarities associated with text classification. As noted by Atanasova et al. (2024) , many explainability techniques are available for text classification. Their survey contributed to the literature by delineating a list of explainability methods used for text classification. Nonetheless, the study did not include rationalization methods and language explanations.

More recently, attention has been focussed on rationalization as a more accessible explainability technique in NLP. Wang and Dou (2022) and Gurrapu et al. (2023) discussed literature around rationalization across various NLP tasks, including challenges and research opportunities in the field. Their work, provides a high-level analysis suitable for a non-technical audience. Similarly, Hartmann and Sonntag (2022) provided a brief overview of methods for learning from human rationales beyond supervised rationalization architectures aiming to inform decision-making for specific use cases. Finally, Wiegreffe and Marasović (2021) identified a list of human-annotated data sets with textual explanations and compared the strengths and shortcomings of existing data collection methodologies. However, it is beyond the scope of this study to examine how these data sets can be used in different rationalization approaches. To the best of our knowledge, no research has been undertaken to survey rationalization methods for text classification.

This survey paper does not attempt to survey all available explainability techniques for text classification comprehensively. Instead, we will compare and contrast state-of-the-art rationalization techniques and their evaluation metrics, providing an easy-to-digest entry point for new researchers in the field. In summary, the objectives of this survey are to:

1. Study and compare different rationalization methods;

2. Compile a list of rationale-annotated data sets for text classification;

3. Describe evaluation metrics for assessing the quality of machine-generated rationales; and

4. Identify knowledge gaps that exist in generating and evaluating rationales.

From January 2007 to December 2023, our survey paper's articles were retrieved from Google Scholar using the keywords “rationales,” “natural language explanations,” and “rationalization.” We have included 88 peer-reviewed publications on NLP and text classification from journals, books, and conference proceedings from venues such as ACL, EMNLP, LREC, COLING, NAACL, AAAI, and NeurIPS.

Figure 2 reveals that there has been a shared increase in the number of research articles on rationalization for explainable text classification since the publication of the first rationalization approach by Lei et al. (2016) . Similarly, the number of research articles on XAI has doubled yearly since 2016. While the number of articles on rationalization peaked in 2021 and has slightly dropped since then to reach 13 articles in 2023, the number of publications on XAI has kept growing steadily. It is important to note that articles published before 2016 focus on presenting rationale-annotated datasets linked to learning with rationales research instead of rationalization approaches within the XAI field.

Figure 2 . Evolution of the number of peer-reviewed publications on rationalization for text classification (bar chart, left y-axis) and XAI (line chart, right y-axis) from 2007 to 2023.

This survey article is organized as follows: Section 2 describes extractive and abstractive rationalization approaches. Section 3 compiles a list of rationale-annotated data sets for text classification. Section 4 outlines evaluation metrics proposed to evaluate and compare rationalization methods. Finally, Section 5 discusses challenges, points out gaps and presents recommendations for future research on rationalization for explainable text classification.

2 Rationalization methods for text classification

We now formalize extractive and abstractive rationalization approaches and compare them in the context of text classification. We define a standard text classification in which we are given an input sequence x = [ x 1 , x 2 , x 3 , …, x l ], where x i is the i -th word of the sequence, and l is the sequence length. The learning problem is to assign the input sequence x to one or multiple labels in y ∈{1, …, c }, where c is the number of classes.

Figure 3 presents an overview of rationalization methods for producing extractive and abstractive rationales. While extractive rationalization models can be categorized into extractive or attention-based methods, abstractive rationalization models can be classified into generative and text-to-text methods. Finally, the component of both extractive and abstractive methods can be trained either using multi-task learning or independently as pipelined architecture.

Figure 3 . Overview of extractive and abstractive rationalization approaches in explainable text classification.

2.1 Extractive rationalization

In extractive rationalization, the goal is to make a text classifier explainable by uncovering parts of the input sequence that the prediction relies on the most ( Lei et al., 2016 ). To date, researchers have proposed two approaches for extractive rationalization for explainable text classification: (i) extractive methods, which first extract evidence from the original text and then make a prediction solely based on the extracted evidence ( Lei et al., 2016 ; Jain et al., 2020 ; Arous et al., 2021 ), and (ii) attention-based methods, which leverage the self-attention mechanism to show the importance of words through their attention weights ( Bao et al., 2018 ; Vashishth et al., 2019 ; Wiegreffe and Pinter, 2019 ).

Table 1 presents an overview of the current techniques for extractive rationalization, where we specify methods, learning approaches taken and their most influential references.

Table 1 . Overview of common approaches for extractive rationalization.

2.1.1 Extractive methods

Most research on extractive methods has been carried out using an encoder-decoder framework ( Lei et al., 2016 ; DeYoung et al., 2020 ; Arous et al., 2021 ). The encoder enc ( x ) works as a tagging model, where each word in the input sequence receives a binary tag indicating whether it is included in the rationales r ( Zaidan et al., 2007 ). The decoder dec ( x, r ) then accepts only the input highlighted as rationales and maps them to one or more target categories ( Bao et al., 2018 ).

The selection of words is performed by an encoder , which is a parameterized mapping enc ( x ) that extracts rationales from input sequences as r = { x i | z i = 1, x i ∈ x }, where z i ∈{0, 1} is a binary tag that indicates whether the word x i is selected or not. In an extractive setting, the rationale r must include only a few words or sentences, and dec ( enc ( x, r )) should result in nearly the same target vector as the original input when passed through the decoder dec ( x ) ( Otter et al., 2020 ; Wang and Dou, 2022 ).

2.1.1.1 Multi-task models

Lei et al. (2016) pioneered the idea of extracting rationales using the encoder-decoder architecture. They proposed utilizing two models and training them jointly to minimize a cost function composed of a classification loss and sparsity-inducing regularization, responsible for keeping the rationales short and coherent. They identified rationales within the input text by assigning a binary Bernoulli variable to each word. Unfortunately, minimizing the expected cost was challenging since it involved summing over all possible choices of rationales in the input sequence. Consequently, they suggested training these models jointly via REINFORCE-based optimization ( Williams, 1992 ). REINFORCE involves sampling rationales from the encoder and training the model to generate explanations using reinforcement learning. As a result, the model is rewarded for producing rationales that align with desiderata defined in its cost function ( Zhang et al., 2021b ).

The key components of the solution proposed by Lei et al. (2016) are binary latent variables and sparsity-inducing regularization. As a result, their solution is marked by non-differentiability. Bastings et al. (2019) proposed to replace the Bernoulli variables with rectified continuous random variables, amenable for reparameterization and for which gradient estimation is possible without REINFORCE. Along the same lines, Madani and Minervini (2023) used Adaptive Implicit Maximum Likelihood ( Minervini et al., 2023 ), a recently proposed low-variance and low-bias gradient estimation method for discrete distribution to back-propagate through the rationale extraction process. Paranjape et al. (2020) emphasized the challenges around the sparsity-accuracy trade-off in norm-minimization methods such as the ones proposed by Lei et al. (2016) and Bastings et al. (2019) . In contrast, they showed that it is possible to better manage this trade-off by optimizing a bound on the Information Bottleneck objective ( Mukherjee, 2019 ) using the divergence between the encoder and a prior distribution with controllable sparsity levels.

Over the last 15 years, research on learning with rationales has established that incorporating human explanations during model training can improve performance and robustness against spurious correlations ( Zaidan et al., 2007 ; Strout et al., 2019 ). Nonetheless, studies on explainability started addressing how human rationales can also help to enhance the quality of explanations for different NLP tasks ( Strout et al., 2019 ; Arous et al., 2021 ) only in the past 4 years.

To determine the impact of a supervised approach for extractive rationalization, DeYoung et al. (2020) adapted the implementation of Lei et al. (2016) , incorporating human rationales during training by modifying the model's cost function. Similarly, Bhat et al. (2021) developed a multi-task teacher-student framework based on self-training language models with limited task-specific labels and rationales. It is important to note that in the variants of the encoder-decoder architecture using human rationales, the final cost function is usually a composite of the classification loss, regularizers on rationale desiderata, and the loss over rationale predictions ( DeYoung et al., 2020 ; Gurrapu et al., 2023 ).

One of the main drawbacks of multi-task learning architectures for extractive rationales is that it is challenging to train the encoder and decoder jointly under instance-level supervision ( Zhang et al., 2016 ; Jiang et al., 2018 ). As described before, these methods sample rationales using regularization to encourage sparsity and contiguity and make it necessary to estimate gradients using either the REINFORCE method ( Lei et al., 2016 ) or reparameterized gradients ( Bastings et al., 2019 ). Both techniques complicate training and require careful hyperparameter tuning, leading to unstable solutions ( Jain et al., 2020 ; Kumar and Talukdar, 2020 ).

Furthermore, recent evidence suggests that multi-task rationalization models may also incur what is called the degeneration problem, where they produce nonsensical rationales due to the encoder overfitting to the noise generated by the decoder ( Madsen et al., 2022 ; Wang and Dou, 2022 ; Liu et al., 2023 ). To tackle this challenge, Liu et al. (2022) introduced a Folded Rationalization approach that folds the two stages of extractive rationalization models into one using a unified text representation mechanism for the encoder and decoder. Using a different approach, Jiang et al. (2023) proposed the YOFO (You Only Forward Once), a simplified single-phase framework with a pre-trained language model to perform prediction and rationalization. It is essential to highlight that rationales extracted using the YOFO framework aim only to support predictions and are not used directly to make model predictions.

2.1.1.2 Pipelined models

Pipelined models are a simplified version of the encoder-decoder architecture in which, first, the encoder is configured to extract the rationales. Then, the decoder is trained separately to perform prediction using only rationales ( Zhang et al., 2016 ; Jain et al., 2020 ). It is important to note that no parameters are shared between the two models and that rationales extracted based on this approach have been learned in an unsupervised manner since the encoder does not have access to human rationales during training.

To avoid the complexity of training a multi-task learning architecture, Jain et al. (2020) introduced FRESH (Faithful Rationale Extraction from Saliency tHresholding). Their scheme proposed using arbitrary feature importance scores to identify the rationales within the input sequence. An independent classifier is then trained exclusively on snippets the encoder provides to predict target labels. Similarly, Chrysostomou and Aletras (2022) proposed a method that also uses gradient-based scores as the encoder. However, their method incorporated additional constraints regarding length and contiguity for selecting rationales. Their work shows that adding these additional constraints can enhance the coherence and relevance of the extracted rationales, ensuring they are concise and contextually connected, thus improving the understanding and usability of the model in real-world applications.

Going beyond feature importance scores, Jiang et al. (2018) suggested using a reinforcement learning method to extract rationales using a reward function based on latent variables to define the extraction of phrases and classification labels. Their work indicates that reinforcement can optimize the rationale selection process, potentially leading to more accurate explanations by adjusting strategies based on feedback to maximize the reward function. Along the same lines, Guerreiro and Martins (2021) developed SPECTRA (SparsE StruCtured Text Rationalization), a framework based on LP-SparseMAP ( Niculae and Martins, 2020 ). Their method provided a flexible, deterministic and modular rationale extraction process based on a constrained structured prediction algorithm. It is important to note that incorporating a deterministic component can eventually boost the consistency and predictability of the extracted rationales, improving the reliability and reproducibility of explanations across different datasets and applications.

Simplifying the encoder-decoder architecture in extractive rationalization models might enhance its use in explainable NLP systems ( Jain et al., 2020 ; Wang and Dou, 2022 ). This simplification can lead to more computationally efficient models, broadening their applicability and accessibility in various real-world scenarios.

Recently, there has been increasing interest in leveraging Large Language Models (LLMs) for extractive rationalization, owing to their ability to efficiently process and distill critical information from large text corpora ( Wang and Dou, 2022 ; Gurrapu et al., 2023 ). The evidence reviewed here suggests that rationalization models might improve performance by prompting language models in a few-shot manner, with rationale-augmented examples. Using this approach, Chen et al. (2023) introduced ZARA, an approach for data augmentation and extractive rationalization using transformer-based models ( Vaswani et al., 2017 ) such as RoBERTa ( Liu et al., 2019b ), DeBERTa ( He et al., 2020 ), and BART ( Lewis et al., 2020 ). Along the same lines, Zhou et al. (2023) presented a two-stage few-shot learning method that first generates rationales using GPT-3 ( Brown et al., 2020 ), and then fine-tunes a smaller rationalization model, RoBERTa, with generated explanations. It is important to consider a few challenges of using LLMs for rationalization models, including high computational demands and the potential for ingrained biases that can skew language explanations ( Zhao et al., 2023 ).

Even though extractive rationalization may be a crucial component of NLP systems as it enhances trust by providing human-understandable explanations, far too little attention has been paid to its use in real-world applications ( Wang and Dou, 2022 ; Kandul et al., 2023 ). ExClaim is a good illustration of using extractive rationalization in a high-stake domain. Gurrapu et al. (2022) introduced ExClaim to provide an explainable claim verification tool for use in the legal sector based on extractive rationales that justify verdicts through natural language explanations. Similarly, Mahoney et al. (2022) presented an explainable architecture based on extractive rationales that explain the results of a machine learning model for classifying legal documents. Finally, Tornqvist et al. (2023) proposed a pipelined approach for extractive rationalization to provide explanations for an automatic grading system based on a transformer-based classifier and post-hoc explanability methods such as SHAP ( Lundberg and Lee, 2017 ) and Integrated Gradients ( Sundararajan et al., 2017 ).

2.1.2 Attention-based methods

Attention models have not only resulted in impressive performance for text classification ( Vaswani et al., 2017 ), but are also suitable as a potential explainability technique ( Vashishth et al., 2019 ; Wiegreffe and Pinter, 2019 ). In particular, the attention mechanism has been previously used to identify influential tokens for the prediction task by providing a soft score over the input units ( Bahdanau et al., 2015 ).

Researchers have drawn inspiration from the model architecture from Jain and Wallace (2019) for text classification. For a given input sequence x , each token is represented by its D -dimensional embedding to obtain x e ∈ ℝ D × d . Next, a bidirectional recurrent neural network (Bi-RNN) encoder is used to obtain an m -dimensional contextualized representation of tokens: h = E n c ( x e ) ∈ ℝ D × m . Finally, the additive formulation of attention proposed by Bahdanau et al. (2015) ( W ∈ ℝ D × D , b, c ∈ ℝ D are parameters of the model) is used for computing weights α i for all tokens defined as in Equation 1 :

The weighted instance representation h α = ∑ i = 1 T α i h i is fed to a dense layer and followed by a softmax function to obtain prediction ỹ = σ ( D e c ( h α ) ) ∈ ℝ | c | where | c | denotes the label set size. Finally, a heuristic strategy must be applied to map attention scores to discrete rationales. Examples include selecting spans within a document based on their total score (sum of their tokens' importance scores) or picking the top-k tokens with the highest attention scores ( Jain et al., 2020 ).

2.1.2.1 Soft-scores models

Some studies have proposed using variants of attention ( Bahdanau et al., 2015 ) to extract rationales in an unsupervised manner. For explainable text classification, Wiegreffe and Pinter (2019) investigated a model that passes tokens through a BERT model ( Devlin et al., 2019 ) to induce contextualized token representations that are then passed to a bidirectional LSTM ( Hochreiter and Schmidhuber, 1997 ). For soft-score features, they focused attention on the contextualized representation. Similarly, Vashishth et al. (2019) analyzed the attention mechanism on a more diverse set of NLP tasks and assessed how attention enables interpretability through manual evaluation.

Bao et al. (2018) extended the unsupervised approach described above by learning a mapping from human rationales to continuous attention. Like the supervised approach for extractive methods, they developed a model to map human rationales onto attention scores to provide richer supervision for low-resource models. Similarly, Strout et al. (2019) showed that supervising attention with human-annotated rationales can improve both the performance and explainability of results of a classifier based on Convolutional Neural Networks (CNNs; Lai et al., 2015 ). In the same vein, Kanchinadam et al. (2020) suggested adding a lightweight attention mechanism to a feed-forward neural network classifier and training them using human-annotated rationales as additional feedback.

Even though these are promising methods for extracting rationales, they require access to a significant number of rationale-annotated instances, which might be impractical for domain-specific applications where expert annotators are rare and constrained for time ( Vashishth et al., 2019 ; Kandul et al., 2023 ). Consequently, Zhang et al. (2021a) proposed HELAS (Human-like Explanation with Limited Attention Supervision). This approach requires a small proportion of documents to train a model that simultaneously solves the text classification task while predicting human-like attention weights. Similarly, Arous et al. (2021) introduced MARTA, a Bayesian framework based on variational inference that jointly learns an attention-based model while injecting human rationales during training. It is important to note that both approaches achieve state-of-the-art results while having access to human rationales for less than 10% of the input documents.

While attention mechanisms have been used for extractive rationalization, their effectiveness as a stand alone explainability method is debated ( Burkart and Huber, 2021 ; Niu et al., 2021 ). Data from several studies suggest that attention weights might misidentify relevant tokens in their explanations, or they are often uncorrelated with the importance score measured by other explainability methods ( Jain and Wallace, 2019 ; Bastings and Filippova, 2020 ). This uncertainty has significantly undermined the use of attention-based methods, as they can provide a false sense of understanding of the model's decision-making process, potentially leading to a misguided trust in the NLP system's capabilities and an underestimation of its limitations ( Kandul et al., 2023 ; Lyu et al., 2024 ).

2.2 Abstractive rationale generation

In abstractive rationalization, the aim is to generate natural language explanations to articulate the model's reasoning process describing why an input sequence was mapped to a particular target vector. Abstractive rationales may involve synthesizing or paraphrasing information rather than directly extracting snippets from the input text ( Liu et al., 2019a ; Narang et al., 2020 ).

Although extractive rationales are very useful to understand the inner workings of a text classifier, there is a limitation when employing them in tasks that should link commonsense knowledge information to decisions, such as natural language inference (NLI), question-answering, and text classification ( Camburu et al., 2018 ; Rajani et al., 2019 ). In such cases, rather than extracting relevant words from the input sequence, it is more desirable to provide a more synthesized and potentially insightful overview of the model's decision-making, often resembling human-like reasoning ( Liu et al., 2019a ; Narang et al., 2020 ).

There are two main approaches currently being adopted in research into abstractive rationalization: (i) text-to-text methods, which rely on sequence-to-sequence translation models such as the Text-to-Text Transfer Transformer (T5) framework proposed by Raffel et al. (2020) including both the label and the explanation at the same time, and (ii) generative methods, which first generate a free-form explanation and then makes a prediction based on the produced abstractive rationale ( Zhou et al., 2020 ). Table 2 presents an overview of the methods used to produce abstractive rationales and their representative references.

Table 2 . Overview of common approaches for abstractive rationale generation.

It is important to note that a relatively small body of literature is concerned with abstractive rationalization for explainable text classification. Abstractive rationales are used less frequently than extractive rationales primarily due to the higher complexity and technical challenges in generating coherent, accurate, and relevant synthesized explanations ( Madsen et al., 2022 ; Ji et al., 2023 ). Consequently, most of the studies on abstractive rationalization have been based on supervised methods, where human explanations are provided during the model's training ( Liu et al., 2019a ; Zhou et al., 2020 ).

2.2.1 Text-to-text methods

A text-to-text model follows the sequence-to-sequence (seq2seq) framework ( Sutskever et al., 2014 ), where it is fed a sequence of discrete tokens as input and produces a new sequence of tokens as output. Using this approach, researchers have leveraged the T5 framework to train a joint model designed to generate explanations and labels simultaneously ( Raffel et al., 2020 ). Consequently, a model is fit to maximize the following conditional likelihood of the target label y and explanations e given the input text x as defined in Equation 2 :

2.2.1.1 Multi-task models

Text-to-text methods for generating abstractive rationales leverage the text-to-text framework proposed by Raffel et al. (2020) to train language models to output natural text explanations alongside their predictions. A study by Narang et al. (2020) showed that their WT5 model (T5 models using “base” and “11B” configurations; Raffel et al., 2020 ) achieved state-of-the-art results with respect to the quality of explanations and classification performance, when having access to a relatively large set of labeled examples. Finally, they also claimed that their WT5 model could help transfer a model's explanation capabilities across different data sets.

Similarly, Jang and Lukasiewicz (2021) conducted experiments evaluating abstractive rationales generated by a T5-base model for text classification and NLI. Nevertheless, their work emphasized the need to reduce the volume of rationale-annotated data and the computational requirements required to train these models to produce comprehensive and contextually appropriate rationales.

Text-to-text models have shown promising results for improving the understanding of classification models and increasing the prediction performance using explanations as additional features ( Gilpin et al., 2018 ; Danilevsky et al., 2020 ). However, their training requires a large number of human-annotated rationales. This property precludes the development of free-text explainable models for high-stake domains where rationale-annotated data sets are scarcely available ( Jang and Lukasiewicz, 2021 ).

2.2.2 Generative methods

Researchers investigating generative methods have utilized a generator-decoder framework ( Camburu et al., 2018 ; Rajani et al., 2019 ), which is similar to the encoder-decoder used for extractive rationalization. The generator gen ( x ) works as a seq2seq model where each input sequence is mapped onto a free-form explanation ( Zhou et al., 2020 ). The decoder dec ( x ) then takes the abstractive rationale to predict the target vector ( Jang and Lukasiewicz, 2021 ).

By using the multiplication law of conditional probability, we can decompose Equation (3) and formulate the training of generative methods as Zhou et al. (2020) :

An explanation generator model gen ( x ) that parameterizes p ( e i | x i ) takes an input sequence x and generates a corresponding natural language explanation e . As mentioned, the abstractive rationale might not be found in the input sequence x ( Zhou et al., 2020 ). The decoder dec ( x, e ) is an augmented prediction model, which parameterizes p ( y i | x i , e i ) and takes an input sequence x and an explanation e to assign a target vector y ( Rajani et al., 2019 ; Atanasova et al., 2020 ).

A significant advantage of generative methods for abstractive rationalization is that they require significantly fewer human-annotated examples for training an explainable text classification model than text-to-text methods. Due to their flexibility in creating new content, generative methods allow for a broader range of expressive and contextually relevant rationales that can closely mimic human-like explanations ( Liu et al., 2019a ; Zhou et al., 2020 ).

2.2.2.1 Pipelined models

As with extractive methods, pipelined models for abstractive rationalization simplify the generator-decoder architecture. Both modules are trained independently, with no parameters shared between the two models. Kumar and Talukdar (2020) proposed a framework where a pre-trained language model based on the GPT-2 architecture ( Radford et al., 2019 ) is trained using a causal language modeling loss (CLM). An independent RoBERTa-based ( Liu et al., 2019b ) classifier is then fit on the abstractive rationales to predict target labels. Similarly, Zhao and Vydiswaran (2021) introduced LiREX, a framework also based on a GPT-2-based generator and a decoder leveraging RoBERTa. However, this framework included an additional component at the start of the pipeline that first extracts a label-aware token-level extractive rationale and employs it to generate abstractive explanations. Due to the possibility of generating label-aware explanations, LiREX is especially suitable for multi-label classification problems.

2.2.2.2 Multi-task models

Drawing inspiration from the work of Camburu et al. (2018) on abstractive rationalization for explainable NLI, Zhou et al. (2020) developed the ELV (Explanations as Latent Variables) framework. They used a variational expectation-maximization algorithm ( Palmer et al., 2005 ) for optimization where an explanation generation module and an explanation-augmented BERT module are trained jointly. They considered natural language explanations as latent variables that model the underlying reasoning process of neural classifiers. Since training a seq2seq model to generate explanations from scratch is challenging, they used UniLM ( Dong et al., 2019 ), a pre-trained language generation model, as the generation model in their framework. Similarly, Li et al. (2021) proposed a joint neural predictive approach to predict and generate abstractive rationales and applied it to English and Chinese medical documents. As generators, they used the large version of T5 (T5 large; Raffel et al., 2020 ) and its multilingual version, mT5 ( Xue et al., 2021 ). For classification, they applied ALBERT ( Lan et al., 2019 ) and RoBERTa ( Liu et al., 2019b ) on the English and Chinese data sets, respectively. Even though they found that the multi-task learning approach boosted model explainability, the improvement in their experiments was not statistically significant.

A few studies have shown that generative methods sometimes fail to build reliable connections between abstractive rationales and predicted outcomes ( Carton et al., 2020 ; Wiegreffe et al., 2021 ). Therefore, there is no guarantee that the generated explanations reflect the decision-making process of the prediction model ( Tan, 2022 ). To generate faithful explanations, Liu et al. (2019a) suggested using an explanation factor to help build stronger connections between explanations and predictions. Their Explanation Factor (EF) considers the distance between the generated and the gold standard rationales and the relevance between the abstractive rationales and the original input sequence. Finally, they included EF in the objective function and jointly trained the generator and decoder to achieve state-of-the-art results for predicting and explaining product reviews.

New findings amongst abstractive rationalization provide further evidence that models are prone to hallucination ( Kunz et al., 2022 ; Ji et al., 2023 ). In explainable text classification, hallucination refers to cases where a model produces factually incorrect or irrelevant rationales, thus impacting the reliability and trustworthiness of these explanations ( Zhao et al., 2023 ). Even though most evaluation metrics punish hallucination and try to mitigate it during training, the irrelevant rationales included might add helpful information for the classification step and, therefore, be used regardless. This phenomenon can mislead users about the model's decision-making process, undermining the credibility of NLP systems and posing challenges for its practical application in scenarios requiring high accuracy and dependability ( Wang and Dou, 2022 ; Ji et al., 2023 ).

Zero-shot approaches are increasingly relevant in NLP as they allow models to process language tasks they have not been explicitly trained on, enhancing their adaptability as part of real-world solutions where training data is not necessarily available ( Meng et al., 2022 ). Even though there is a relatively small body of literature that is concerned with zero-shot rationalization approaches for explainable text classification, studies such as that conducted by Kung et al. (2020) and Lakhotia et al. (2021) have shown that zero-shot rationalization models achieve comparable performance without any supervised signal. Nevertheless, a significant challenge is the model's ability to produce relevant rationales for unseen classes, as it must extrapolate from learned concepts without direct prior knowledge ( Lyu et al., 2021 ). This capability requires understanding abstract and transferable features across different contexts, difficulting the training and deployment of these rationalization models ( Wei et al., 2021 ; Meng et al., 2022 ). It is important to note that, if successful, they can enhance the scalability of NLP systems by making them capable of analyzing data from various domains without needing extensive retraining ( Kung et al., 2020 ; Yuan et al., 2024 ).

3 Rationale-annotated datasets

During the last 15 years, there has been an increase in the volume of rationale-annotated data available, boosting progress on designing more explainable classifiers and facilitating the evaluation and benchmarking of rationalization approaches ( DeYoung et al., 2020 ; Wang and Dou, 2022 ).

Table 3 describes each rationale-annotated dataset for text classification in terms of their domain, the annotation procedure used to collect the human explanations (indicated as “author” or “crowd” for crowd-annotated), their number of instances (input-label pairs), their publication year and the original paper where they were presented. Moreover, it includes links to each dataset (when available), providing direct access for further exploration and detailed analysis.

Table 3 . Comparison of rationale-annotated datasets for text classification.

Incorporating human rationales during training of supervised learning models can be traced back to the work of Zaidan et al. (2007) , where a human teacher highlighted text spans in a document to improve model performance. Their MovieReviews(v.1.0) corpus is the first rationale-annotated dataset for text classification, including 1,800 positive/negative sentiment labels on movie reviews.

Table 3 shows that the dominant collection paradigm is via crowd sourcing platforms. A critical bottleneck of rationale generation is the insufficient domain-specific rationale-annotated data ( Lertvittayakumjorn and Toni, 2019 ). Gathering enough (input, label, and human rationales ) triples from potential end-users is essential as it provides rationalization models with a reference for what constitutes a meaningful and understandable explanation from a human perspective ( Strout et al., 2019 ; Carton et al., 2020 ; DeYoung et al., 2020 ). Rationale-annotated data is critical in real-world applications, where the alignment of machine-generated rationales with human reasoning greatly enhances the model's transparency, trustworthiness, and acceptance by users in practical scenarios ( Wang and Dou, 2022 ; Gurrapu et al., 2023 ).

Creating benchmark data sets with human annotations is essential for training and comparing rationalization models, as they provide a standardized resource to evaluate the effectiveness, accuracy, and human-likeness of model-generated explanations ( Jacovi and Goldberg, 2021 ; Wang and Dou, 2022 ). Such benchmarks facilitate consistent, objective comparison across different models, fostering advancements in the field by highlighting areas of strength and opportunities for improvement in aligning machine-generated explanations with human reasoning and understanding ( Kandul et al., 2023 ; Lyu et al., 2024 ). The task of extractive rationalization was surveyed by DeYoung et al. (2020) , who proposed the ERASER (Evaluating Rationales And Simple English Reasoning) benchmark spanning a range of NLP tasks. These data sets, including examples for text classification such as MovieReviews(v.2.0) and FEVER, have been repurposed from pre-existing corpora and augmented with labeled rationales ( Zaidan et al., 2007 ; Thorne et al., 2018 ). More recently, Marasović et al. (2022) introduced the FEB benchmark containing four English data sets for few-shot rationalization models, including the SBIC corpus for offensiveness classification.

Questions have been raised about using human-annotated rationales for training and evaluating rationalization models since they are shown to be quite subjective ( Lertvittayakumjorn and Toni, 2019 ; Carton et al., 2020 ). Most published studies failed to specify information about the annotators, such as gender, age, or ethnicity. Jakobsen et al. (2023) makes an essential contribution by being the first dataset to include annotators' demographics and human rationales for sentiment analysis. Diversity in collecting human rationales is crucial to the development of universally understandable and reliable models, enhancing their applicability and acceptance across a broad spectrum of stakeholders and scenarios ( Tan, 2022 ; Yao et al., 2023 ).

Finally, different methods have been proposed to collect human rationales for explainable text classification. On the one hand, in some studies (e.g., Zaidan et al., 2007 ), annotators were asked to identify the most important phrases and sentences supporting a label. On the other hand, in the work of Sen et al. (2020) , for example, all sentences relevant to decision-making were identified. Even though these approaches seem similar, they might lead to substantially different outcomes ( Hartmann and Sonntag, 2022 ; Tan, 2022 ). Documentation and transparency in the annotation of human rationales are essential as they provide clear insight into the reasoning process and criteria used by human annotators, ensuring replicability and trustworthiness in the model evaluation process ( Carton et al., 2020 ). This detailed documentation is crucial for understanding potential biases and the context under which these rationales were provided, thereby enhancing the credibility and generalizability of the rationalization models.

4 Evaluation metrics

The criteria for evaluating the quality of rationales in explainable text classification are not universally established. Generally, evaluation approaches fall into two categories: (i) proxy-based , where rationales are assessed based on automatic metrics that attempt to measure different desirable properties ( Carton et al., 2020 ; DeYoung et al., 2020 ), and (ii) human-grounded , where humans evaluate rationales in the context of a specific application or a simplified version of it ( Doshi-Velez and Kim, 2017 ; Lertvittayakumjorn and Toni, 2019 ).

Table 4 summarizes the categories for rationale evaluation, including metrics and their most relevant references.

Table 4 . Overview of evaluation metrics for rationale's quality.

4.1 Proxy-based

Plausibility in rationalization for text classification refers to the extent to which explanations provided by a model align with human intuition and understanding ( DeYoung et al., 2020 ; Wiegreffe et al., 2021 ). Plausible explanations enhance the trust and credibility of classifiers, as they are more likely to be understood and accepted by end-users, particularly those without technical expertise ( Doshi-Velez and Kim, 2017 ; Hase and Bansal, 2022 ; Atanasova et al., 2024 ). DeYoung et al. (2020) proposed evaluating plausibility using Intersection-over-Union at the token level to derive token-level precision, recall, and F1 scores. Several studies have followed a similar evaluation approach for extractive rationalization models ( Paranjape et al., 2020 ; Guerreiro and Martins, 2021 ; Chan A. et al., 2022 ), while others have explored using phrase-matching metrics such as SacreBLEU and METEOR ( Jang and Lukasiewicz, 2021 ) for evaluating abstractive rationales. In the case of attention-based methods that perform soft selection, DeYoung et al. (2020) suggested measuring plausibility using the Area Under the Precision-Recall Curve (AUPRC) constructed by sweeping a threshold over token scores ( DeYoung et al., 2020 ; Chan A. et al., 2022 ).

While plausibility is important for rationalization models, much of the literature acknowledges that generating plausible rationales is not enough ( Doshi-Velez and Kim, 2017 ; Arrieta et al., 2020 ; Danilevsky et al., 2020 ). Previous research has established that it is crucial to ensure that the rationales also reflect the actual reasoning processes of the model rather than being superficial or misleading ( Belle and Papantonis, 2021 ; Jacovi and Goldberg, 2021 ). Faithfulness refers to the degree to which the generated rationales accurately represent the internal decision-making process of the model. DeYoung et al. (2020) proposed two automatic metrics for assessing faithfulness by measuring the impact of perturbing or erasing snippets within language explanations. First, comprehensiveness captures the extent to which all relevant features for making a prediction were selected as rationales. Second, sufficiency assesses whether the snippets within rationales are adequate for a model to make a prediction. Using this approach, researchers have established that a faithful rationale should have high comprehensiveness and sufficiency ( Zhang et al., 2021a ; Chan A. et al., 2022 ).

Supporting this view, Carton et al. (2020) introduced the term fidelity to refer jointly to sufficiency and comprehensiveness. According to their findings, a rationale can contain many tokens irrelevant to the prediction while still having high comprehensiveness and low sufficiency. Consequently, they introduced the idea of fidelity curves to assess rationale irrelevancy by looking at how sufficiency and comprehensiveness degrade as tokens are randomly occluded from a language explanation. There is a consensus among researchers and practitioners that this level of authenticity in explanations is crucial for users to scrutinize NLP decisions, particularly in high-stake domains where understanding the model's reasoning is paramount ( Miller, 2019 ; Tjoa and Guan, 2020 ; Bibal et al., 2021 ).

Robustness refers to the model's ability to consistently provide reliable rationales across various inputs and conditions ( Gunning et al., 2019 ; Arrieta et al., 2020 ; Lyu et al., 2024 ). Robustness is crucial for explainable text classification as it ensures dependability and generalizability of the explanations, particularly in real-world applications where data variability and unpredictability are common ( Belle and Papantonis, 2021 ; Hartmann and Sonntag, 2022 ). Most researchers investigating robustness in rationalization models have utilized adversarial examples to evaluate the model's rationales to remain trustworthy and reliable in potentially deceptive environments ( Zhang et al., 2020 ; Liang et al., 2022 ). Using this approach, Chen H. et al. (2022) assessed the model's robustness by measuring performance on challenge datasets where human-annotated edits to inputs that can change classification labels, are available. Similarly, Ross et al. (2022) proposed assessing robustness by testing whether rationalization models are invariant to adding additional sentences and remain consistent with their predictions. Data from both studies suggest that rationalization models can improve robustness. However, leveraging human rationales as extra supervision does not always translate to more robust models.

It is important to note that most rationale evaluation research has focused on extractive rationalization models ( Carton et al., 2020 ; Hase and Bansal, 2020 ). Assessing abstractive rationales for explainable text classification presents several unique challenges. First, the subjective nature of abstractive rationales makes standardization of evaluation metrics, such as plausibility difficult, as these rationales do not necessarily align with references of the original input text ( Camburu et al., 2020 ; Zhao and Vydiswaran, 2021 ). Second, ensuring faithfulness and robustness of abstractive rationales is complex, as they involve generating new text that may not directly correspond to specific input features, making it challenging to determine whether the rationale reflects the model's decision-making reliably ( Dong et al., 2019 ; Zhou et al., 2020 ). These challenges highlight the need for innovative and adaptable evaluation frameworks that can effectively capture the multifaceted nature of abstractive rationales in explainable NLP systems.

4.2 Human-grounded

Even though the vast majority of research on rationale evaluation has been proxy-based, some studies have begun to examine human-grounded evaluations for explainable text classification ( Mohseni et al., 2018 ; Ehsan et al., 2019 ). Nevertheless, to our knowledge, there is no published research on human-grounded methods using domain experts in the same target application. Instead, we have found some studies conducting simpler human-subject experiments that maintain the essence of the target application.

According to Ehsan et al. (2019) , rationale understandability refers to the degree to which a rationale helped an observer understand why a model behaved as it did. They asked participants to rate the understandability of a set of rationales using a 5-point Likert scale. Instead, Lertvittayakumjorn and Toni (2019) used binary forced-choice experiments. As part of their research, humans were presented with pairs of explanations to choose the one they found more understandable.

Finally, researchers have also been interested in measuring simulatability using human-subject simulation experiments. In a qualitative study by Lertvittayakumjorn and Toni (2019) , humans were presented with input-explanation pairs and asked to simulate the model's outcomes correctly. Similarly, Ehsan et al. (2019) assessed simulatability using counterfactual simulation experiments. In this case, observers were presented with input-output-explanation triples and asked to identify what words needed to be modified to change the model's prediction to the desired outcome.

In an investigation into human-grounded metrics for evaluating rationales in text classification, Lertvittayakumjorn and Toni (2019) concluded that experiments and systems utilized to collect feedback on machine-generated rationales lack interactivity. In almost every study, users cannot contest a rationale or ask the system to explain the prediction differently. This view is supported by Ehsan et al. (2019) , who concluded that current human-grounded experiments could only partially assess the potential implications of language explanations in real-world scenarios.

Even though human-grounded evaluation is key in assessing the real-world applicability and effectiveness of rationalization models, it presents several challenges that stem from the inherent subjectivity and variability of human judgment ( Doshi-Velez and Kim, 2017 ; Carton et al., 2020 ). First, the diversity of interpretations among different evaluators can lead to an inconsistent assessment of the quality and relevance of the generated rationales ( Lertvittayakumjorn and Toni, 2019 ; Hase and Bansal, 2020 ). As mentioned before, this diversity is influenced by cultural background, domain expertise, and personal biases, making it difficult to consolidate a standardized evaluation metric ( Mohseni et al., 2018 ; Yao et al., 2023 ). Second, the cognitive load on human evaluators can be significant, especially when dealing with complex classification tasks or lengthy rationales, potentially affecting the consistency and reliability of their judgment ( Tan, 2022 ). Finally, there is the scalability challenge, as human evaluations are time-consuming and resource-intensive, limiting the feasibility of conducting large-scale assessments ( Kandul et al., 2023 ).

5 Challenges and future outlook

In this section, we discuss the current challenges in developing trustworthy rationalization models for explainable text classification and suggest possible approaches to overcome them.

5.1 Rationalization approaches

Extractive and abstractive rationalization approaches have distinct advantages and disadvantages when applied to explainable text classification. Table 5 summarizes the trade-offs of the rationalization methods described in Section 2.

Table 5 . Main advantages and disadvantages of methods for rationale generation.

Extractive rationalization, which involves selecting parts of the input text as justification for the model's decision, boasts the advantage of being directly linked to the original data, often making these explanations more straightforward and more accessible to validate for accuracy ( Wang and Dou, 2022 ; Gurrapu et al., 2023 ). However, this method can be limited in providing context or explaining decisions requiring synthesizing information not explicitly stated in the text ( Kandul et al., 2023 ; Lyu et al., 2024 ). Abstractive rationalization, which generates new text to explain the model's decision, offers greater flexibility and can provide more holistic and nuanced explanations that synthesize various aspects of the input data. This approach can be more intuitive and human-like, enhancing the comprehensibility for end-users ( Li et al., 2021 ; Zini and Awad, 2022 ). Yet, it faces challenges such as the risk of hallucination—producing explanations that are not grounded in the input data—and the complexity of ensuring that these generated explanations are both accurate and faithful to the model's decision-making process ( Liu et al., 2019a ; Hase and Bansal, 2020 ). Therefore, while extractive methods offer reliability and direct traceability, abstractive methods provide richness and depth, albeit with increased challenges in maintaining fidelity and accuracy ( Wiegreffe et al., 2021 ; Yao et al., 2023 ).

The choice between extractive and abstractive rationalization models for explainable text classification largely depends on the specific requirements and constraints of the application ( Wang and Dou, 2022 ; Gurrapu et al., 2023 ). On the one hand, extractive rationalization models are generally more suitable in scenarios where transparency and direct traceability to the original text are paramount. They are ideal when the rationale for a decision needs to be anchored to specific parts of the input text, such as in legal or compliance-related tasks where every decision must be directly linked to particular evidence or clauses ( Bibal et al., 2021 ; Lyu et al., 2024 ). On the other hand, abstractive rationalization models are better suited for scenarios where a more synthesized understanding or a broader context is necessary ( Miller, 2019 ; Kandul et al., 2023 ). They excel in situations where the rationale might involve drawing inferences or conclusions not explicitly stated in the text. Abstractive models are also preferable when the explanation needs to be more accessible to laypersons, as they can provide more natural, human-like explanations ( Amershi et al., 2014 ; Tjoa and Guan, 2020 ).

Even though the decision to use pipelined or multi-task learning models for rationalization depends on the specific goals and constraints, several studies suggest that multi-task learning models perform better for both extractive and abstractive rationalization ( Dong et al., 2019 ; Zhou et al., 2020 ; Li et al., 2021 ; Wang and Dou, 2022 ). Pipelined models are advantageous when each module, rationalization and classification, require specialized handling or when modularity is needed in the system ( Jain et al., 2020 ; Chrysostomou and Aletras, 2022 ). This approach allows for greater flexibility in updating each component independently. However, they can suffer from error propagation where the rationalization can affect the classification ( Kunz et al., 2022 ). In contrast, multi-task learning models are generally more efficient and can offer performance benefits, enabling sharing of insights between tasks. Nevertheless, they may require more training data, more complex hyperparameter tuning and careful balancing of the learning objectives ( Bastings et al., 2019 ; Chan A. et al., 2022 ). Finally, the choice depends on the specific requirements for model performance, the availability of training data, and the need for flexibility in model deployment and maintenance.

Since approaches have been trained and tested on different datasets using a variety of evaluation metrics, we have ranked them based on their reported performance on the MovieReviews ( Zaidan et al., 2007 ), SST ( Socher et al., 2013 ), and FEVER ( Thorne et al., 2018 ) datasets. Table 6 compares the performance of each rationalization approach in terms of its predictive performance and the quality of its produced rationales using sufficiency and comprehensiveness scores. Based on the results reported by the authors, we have categorized the predictive performance into: ✓✓✓—Very good performance, ✓✓—Good performance, and ✓— Performance has potential for improvement. What stands out in this table is the dominance of multi-task methods over pipelined and soft-score approaches in terms of predictive performance and explainability. Our summary shows that supervised multi-task extractive approaches are state-of-the-art for rationalization in terms of predictive performance and rationales' quality, followed by supervised multi-task text-to-text abstractive methods. We refer the reader to bf for details of each rationalization approach's performance.

Table 6 . Summary of the evaluation of each rationalization approach in terms of its predictive capability and the quality of its generated explanations.

Combining extractive and abstractive rationales for explainable text classification represents an innovative approach that harnesses the strengths of both: the direct, evidence-based clarity of extractive rationales and the comprehensive, context-rich insights of abstractive explanations. A recent study by Majumder et al. (2022) introduced RE x C (Extractive Rationales, Natural Language Explanations, and Commonsense), a rationalization framework that explains its prediction using a combination of extractive and abstractive language explanations. RE x C selects a subset of the input sequence as an extractive rationale using an encoder based on the HardKuma distribution ( Bastings et al., 2019 ), passes the selected snippets to a BART-based generator ( Lewis et al., 2020 ), and inputs the abstractive rationales to a decoder that outputs the final prediction. It is essential to highlight that all models are trained jointly, and the supervision comes from the target vectors and human-annotated explanations.

Beyond unimodal rationalization models for explainable text classification, multimodal explanations, which integrate textual, visual, and sometimes structured information, can provide more comprehensive insights into AI models' decision-making processes ( Park et al., 2018 ). Using this approach, Marasović et al. (2020) have produced abstractive rationales for visual reasoning tasks, such as visual-textual entailment, by combining pre-trained language models with object recognition classifiers to provide image understanding at the semantic and pragmatic levels. Along the same lines, Zhang et al. (2024) developed a vision language model to identify emotions in visual art and explain their prediction through abstractive rationales. Recent evidence suggests that multimodal explanations can allow for a deeper understanding of how different types of data can be analyzed to produce more accessible and intuitive explanations, broadening the scope and applicability of rationalization in real-world scenarios ( Chen and Zhao, 2022 ; Ananthram et al., 2023 ; Zhang et al., 2024 ).

5.2 Rationale-annotated data

Generating more rationale-annotated data is crucial for training and evaluating rationalization models, as it provides a rich, diverse foundation for teaching these models how to produce relevant and human-understandable explanations ( Doshi-Velez and Kim, 2017 ; Hase and Bansal, 2020 ). These data sets enhance the model's ability to generate accurate and more contextually appropriate rationales and facilitate more robust and comprehensive evaluation, improving the model's reliability and effectiveness in real-world applications. Even though there has been vast progress since the publication of ERASER ( DeYoung et al., 2020 ) and FEB ( Marasović et al., 2022 ) benchmarks, there is still a lack of rationale-annotated data for text classification. Considering that highlighting human rationales is not significantly more expensive than traditional labeling ( Zaidan et al., 2007 ), the NLP community could move toward methods for collecting labels by annotating rationales. By doing so, we could boost the results of classification and rationalization models ( Arous et al., 2021 ).

However, it is not enough to have more rationale-annotated data. We also need better human rationales. Standardizing methods for collecting rationale-annotated data is pivotal in the development of rationalization models, as it ensures a uniform approach to gathering and interpreting data, crucial for maintaining the quality and consistency of training and evaluation processes ( Wiegreffe et al., 2021 ; Yao et al., 2023 ). Documenting and reporting these procedures is equally important, providing transparency about how the data was annotated and allowing applicability in future research ( Atanasova et al., 2020 ; Li et al., 2021 ). Moreover, reporting and fostering the diversity of the annotators involved is critical. Diversity in demographics, expertise, and cognitive perspectives significantly shape machine-generated rationales ( Jakobsen et al., 2023 ). A comprehensive approach to data annotation is vital to advancing rationalization models that are reliable, effective and ethically sound in their explanations, catering to a broad spectrum of real-world applications and stakeholders.

Further work is needed to establish whether crafting datasets annotated with multimodal explanations can enrich the training and capabilities of rationalization approaches for explainable NLP. Even though preliminary results seem to indicate those visual and textual rationales can indeed provide explanatory strengths ( Chen and Zhao, 2022 ; Ananthram et al., 2023 ), one of the main challenges is the complexity involved in integrating diverse data types to ensure that annotations reflect the interconnectedness of these modalities ( Marasović et al., 2020 ). Moreover, developing robust annotation guidelines that capture the nuances of multimodal interactions is complex and requires interdisciplinary expertise ( Yuan et al., 2024 ; Zhang et al., 2024 ).

Since the reasoning process needed to infer a label is subjective and unstructured, we must develop dynamic, flexible and iterative strategies to collect human rationales ( Doshi-Velez and Kim, 2017 ). Considering that we aim to describe the decision-making process in real-world applications accurately, we could move toward noisy data labeling processes attempting to reflect the annotator's internal decision procedure. To illustrate, if annotators change their minds while highlighting rationales, dynamic approaches should be able to capture these changes so that we can learn from them ( Ehsan et al., 2019 ). This dynamic approach might allow for a more authentic and comprehensive representation of human cognitive processes, enriching the training and evaluation of rationalization models with insights that mirror the nature of real-world human thought and decision-making.

The use of human rationales has been key to the development of explainable text classification models. However, further research should focus on whether humans can provide explanations that can later be used to train rationalization models ( Miller, 2019 ; Tan, 2022 ). We need to acknowledge that human rationales, while a valid proxy mechanism, can only help us to understand the decision-making process of humans partially ( Amershi et al., 2014 ). Consequently, we encourage the NLP community to stop looking at them as another set of uniform labels and embrace their complexity by working collaboratively with researchers in other domains. For instance, to understand whether data sets of human explanations can serve their intended goals in real-world applications, we must connect the broad range of notions around human rationales in NLP with existing psychology and cognitive science literature. A more holistic understanding of human explanations should allow us to decide what kind of explanations are desired for NLP systems and help clarify how to generate and use them appropriately within their limitations.

5.3 Comprehensive rationale evaluation

While significant progress has been made in evaluating rationalization models, areas require improvement to ensure safer and more sustainable evaluation ( Lertvittayakumjorn and Toni, 2019 ; Carton et al., 2020 ). Even though current approaches offer valuable insights, there is a need for evaluation frameworks that can assess the suitability and usefulness of the rationales in diverse and complex real-world scenarios ( Chen H. et al., 2022 ; Hase and Bansal, 2022 ). Additionally, there is a growing need to focus on the ethical implications of rationale evaluation, particularly in sensitive applications ( Atanasova et al., 2023 ; Joshi et al., 2023 ). As a community of researchers and practitioners, we must ensure that the models do not inadvertently cause harm or perpetuate misinformation. Addressing these challenges requires a concerted effort from the XAI community to innovate and collaborate, paving the way for more reliable, fair, and transparent rationalization models in NLP.

We have provided a list of diagnostic properties for assessing rationales. It is important to note that these evaluation metrics have mainly been generated from a developer-based perspective, which has biased their results toward faithful explanations ( Lertvittayakumjorn and Toni, 2019 ; DeYoung et al., 2020 ). Current evaluation approaches are not designed nor implemented considering the perspective of other relevant stakeholders, such as investors, business executives, end-users, and policymakers, among many others. Further work must be done to evaluate rationale quality from a broader perspective, including practical issues that might arise in their implementation for real-world applications ( Tan, 2022 ).

Considering how important language explanations are for building trust with end-users ( Belle and Papantonis, 2021 ), their contribution should also be evaluated in the context of their specific application ( Doshi-Velez and Kim, 2017 ). A lack of domain-specific annotated data is detrimental to developing explainable models for high-stake sectors such as the legal, medical and humanitarian domains ( Jacovi and Goldberg, 2021 ; Mendez et al., 2022 ). As mentioned before, current evaluation methods lack interactivity ( Carton et al., 2020 ). End users or domain experts cannot contest rationales or ask the models to explain them differently, which makes them impossible to validate and deploy in real-world applications. Even though it is beyond the scope of our survey, work needs to be done to develop clear, concise and user-friendly ways of presenting rationales as part of explainable NLP systems ( Hartmann and Sonntag, 2022 ; Tan, 2022 ). Effectively communicated rationales boost user trust and confidence in the system and facilitate a deeper comprehension of the model's decision-making process, leading to more informed and effective use of NLP models.

6 Conclusions

Developing understandable and trustworthy systems becomes paramount as NLP and text classification applications continue to integrate into critical and sensitive applications. The present survey article aimed to examine rationalization approaches and their evaluation metrics for explainable text classification, providing a comprehensive entry point for new researchers and practitioners in the field.

The contrast between extractive and abstractive rationalization highlights distinct strengths and limitations. On the one hand, extractive rationalization approaches link to original data, ensuring reliability and ease of validation. However, they may lack the context or comprehensive insight needed for decision-making. On the other hand, abstractive rationalization models offer the flexibility to produce more intuitive and human-like explanations, which enhance user usability and trust. Nevertheless, they face challenges such as the potential for generating non-factual explanations and the complexity of maintaining plausibility in the decision-making process. Choosing between extractive and abstractive models depends on application-specific needs: extractive models are preferable where direct traceability is crucial, such as legal applications. In contrast, abstractive models are suited for situations requiring broader contextual interpretations.

Despite its challenging nature, the emerging work on rationalization for explainable text classification is promising. Nevertheless, several questions remain to be answered. Further research is required to better understand human rationales, establish procedures for collecting them, and develop accurate and feasible methods for generating and evaluating rationales in real-world applications. We have identified possible directions for future research, which will hopefully extend the work achieved so far.

Author contributions

EM: Conceptualization, Investigation, Methodology, Validation, Visualization, Writing – original draft, Writing – review & editing. VS: Conceptualization, Supervision, Writing – review & editing. RB-N: Conceptualization, Supervision, Writing – review & editing.

The author(s) declare financial support was received for the research, authorship, and/or publication of this article. This study was supported by the Chilean National Agency for Research and Development (Scholarship ID 720210003), whose contribution was essential in conducting this research.

Conflict of interest

VS was employed at ASUS Intelligent Cloud Services.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Aggarwal, C. C., and Zhai, C. (2012). “A survey of text classification algorithms,” in Mining Text Data (Boston, MA: Springer), 163–222. doi: 10.1007/978-1-4614-3223-4_6

Crossref Full Text | Google Scholar

Amershi, S., Cakmak, M., Knox, W. B., and Kulesza, T. (2014). Power to the people: the role of humans in interactive machine learning. Ai Mag . 35, 105–120. doi: 10.1609/aimag.v35i4.2513

Ananthram, A., Winn, O., and Muresan, S. (2023). Feelingblue: a corpus for understanding the emotional connotation of color in context. Trans. Assoc. Comput. Linguist . 11, 176–190. doi: 10.1162/tacl_a_00540

Arous, I., Dolamic, L., Yang, J., Bhardwaj, A., Cuccu, G., and Cudré-Mauroux, P. (2021). “Marta: leveraging human rationales for explainable text classification,” in Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 35 (Burnaby, BC: PKP PS), 5868–5876.

Google Scholar

Arrieta, A. B., Díaz-Rodríguez, N., Del Ser, J., Bennetot, A., Tabik, S., Barbado, A., et al. (2020). Explainable artificial intelligence (XAI): concepts, taxonomies, opportunities and challenges toward responsible AI. Inform. Fus . 58, 82–115. doi: 10.1016/j.inffus.2019.12.012

Atanasova, P., Camburu, O. M., Lioma, C., Lukasiewicz, T., Simonsen, J. G., and Augenstein, I. (2023). “Faithfulness tests for natural language explanations,” in Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) (Toronto, ON: Association for Computational Linguistics), 283–294.

Atanasova, P., Simonsen, J. G., Lioma, C., and Augenstein, I. (2020). “Generating fact checking explanations,” in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (Springer Nature Switzerland), 7352–7364.

Atanasova, P., Simonsen, J. G., Lioma, C., and Augenstein, I. (2024). “A diagnostic study of explainability techniques for text classification,” in Accountable and Explainable Methods for Complex Reasoning over Text (Springer Nature Switzerland), 155–187.

PubMed Abstract | Google Scholar

Bahdanau, D., Cho, K., and Bengio, Y. (2015). “Neural machine translation by jointly learning to align and translate,” in 3rd International Conference on Learning Representations, ICLR 2015 .

Bao, Y., Chang, S., Yu, M., and Barzilay, R. (2018). “Deriving machine attention from human rationales,” in Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (Brussels: Association for Computational Linguistics), 1903–1913.

Bashier, H. K., Kim, M. Y., and Goebel, R. (2020). “RANCC: rationalizing neural networks via concept clustering,” in Proceedings of the 28th International Conference on Computational Linguistics (Barcelona: International Committee on Computational Linguistics), 3214–3224.

Bastings, J., Aziz, W., and Titov, I. (2019). “Interpretable neural predictions with differentiable binary variables,” in Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (Florence: ACL Anthology), 2963–2977.

Bastings, J., and Filippova, K. (2020). “The elephant in the interpretability room: why use attention as explanation when we have saliency methods?,” in Proceedings of the Third BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP (Association for Computational Linguistics), 149–155.

Belle, V., and Papantonis, I. (2021). Principles and practice of explainable machine learning. Front. Big Data 4:688969. doi: 10.3389/fdata.2021.688969

PubMed Abstract | Crossref Full Text | Google Scholar

Bhat, M. M., Sordoni, A., and Mukherjee, S. (2021). Self-training with few-shot rationalization: teacher explanations aid student in few-shot NLU. arXiv preprint arXiv:2109.08259 . doi: 10.48550/arXiv.2109.08259

Bibal, A., Lognoul, M., De Streel, A., and Frénay, B. (2021). Legal requirements on explainability in machine learning. Artif. Intell. Law 29, 149–169. doi: 10.1007/s10506-020-09270-4

Blitzer, J., Dredze, M., and Pereira, F. (2007). “Biographies, bollywood, boom-boxes and blenders: domain adaptation for sentiment classification,” in Proceedings of the 45th Annual Meeting of the Association of Computational Linguistics (Prague: Association for Computational Linguistics), 440–447.

Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D., Dhariwal, P., et al. (2020). Language models are few-shot learners. Adv. Neural Inform. Process. Syst . 33, 1877–1901. doi: 10.48550/arXiv.2005.14165

Burkart, N., and Huber, M. F. (2021). A survey on the explainability of supervised machine learning. J. Artif. Intell. Res . 70, 245–317. doi: 10.48550/arXiv.2011.07876

Camburu, O.-M., Rocktäschel, T., Lukasiewicz, T., and Blunsom, P. (2018). e-SNLI: natural language inference with natural language explanations. Adv. Neural Inform. Process. Syst . 31:1193. doi: 10.48550/arXiv.1812.01193

Camburu, O. M., Shillingford, B., Minervini, P., Lukasiewicz, T., and Blunsom, P. (2020). “Make up your mind! adversarial generation of inconsistent natural language explanations,” in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (Association for Computational Linguistics), 4157–4165.

Carton, S., Mei, Q., and Resnick, P. (2018). “Extractive adversarial networks: high-recall explanations for identifying personal attacks in social media posts,” in Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (Brussels: Association for Computational Linguistics), 3497–3507.

Carton, S., Rathore, A., and Tan, C. (2020). “Evaluating and characterizing human rationales,” in Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) (Association for Computational Linguistics), 9294–9307.

Chan, A., Sanjabi, M., Mathias, L., Tan, L., Nie, S., Peng, X., et al. (2022). “UNIREX: a unified learning framework for language model rationale extraction,” in International Conference on Machine Learning (Baltimore, MD: PMLR), 2867–2889.

Chang, S., Zhang, Y., Yu, M., and Jaakkola, T. (2019). A game theoretic approach to class-wise selective rationalization. Adv. Neural Inform. Process. Syst . 32:12853. doi: 10.48550/arXiv.1910.12853

Chen, H., He, J., Narasimhan, K., and Chen, D. (2022). “Can rationalization improve robustness?,” in 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL 2022 (Seattle, WA: Association for Computational Linguistics (ACL)), 3792–3805.

Chen, H., Zheng, G., and Ji, Y. (2020). “Generating hierarchical explanations on text classification via feature interaction detection,” in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics , 5578–5593.

Chen, S., and Zhao, Q. (2022). “REX: reasoning-aware and grounded explanation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (New Orleans, LA: IEEE), 15586–15595.

Chen, W. L., Yen, A. Z., Huang, H. H., Wu, C. K., and Chen, H. H. (2023). “ZARA: improving few-shot self-rationalization for small language models,” in Findings of the Association for Computational Linguistics: EMNLP 2023 (Singapore: Association for Computational Linguistics), 4682–4693.

Chrysostomou, G., and Aletras, N. (2022). “Flexible instance-specific rationalization of NLP models,” in Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 36 (Burnaby, BC: PKP PS), 10545–10553.

Danilevsky, M., Qian, K., Aharonov, R., Katsis, Y., Kawas, B., and Sen, P. (2020). “A survey of the state of explainable AI for natural language processing,” in Proceedings of the 1st Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 10th International Joint Conference on Natural Language Processing (Suzhou: Association for Computational Linguistics), 447–459.

Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. (2019). “BERT: pre-training of deep bidirectional transformers for language understanding,” in Proceedings of NAACL-HLT (Minneapolis, MN: Association for Computational Linguistics), 4171–4186.

DeYoung, J., Jain, S., Rajani, N. F., Lehman, E., Xiong, C., Socher, R., et al. (2020). “ERASER: a benchmark to evaluate rationalized NLP models,” in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (Association for Computational Linguistics), 4443–4458.

Dong, L., Yang, N., Wang, W., Wei, F., Liu, X., Wang, Y., et al. (2019). Unified language model pre-training for natural language understanding and generation. Adv. Neural Inform. Process. Syst . 32:3197. doi: 10.48550/arXiv.1905.03197

Doshi-Velez, F., and Kim, B. (2017). Towards a rigorous science of interpretable machine learning. arXiv preprint arXiv:1702.08608 . doi: 10.48550/arXiv.1702.08608

Ehsan, U., Tambwekar, P., Chan, L., Harrison, B., and Riedl, M. O. (2019). “Automated rationale generation: a technique for explainable AI and its effects on human perceptions,” in Proceedings of the 24th International Conference on Intelligent User Interfaces (Association for Computing Machinery), 263–274.

Gilpin, L. H., Bau, D., Yuan, B. Z., Bajwa, A., Specter, M., and Kagal, L. (2018). “Explaining explanations: an overview of interpretability of machine learning,” in 2018 IEEE 5th International Conference on Data Science and Advanced Analytics (DSAA) (Turin: IEEE), 80–89.

Guerreiro, N. M., and Martins, A. F. (2021). “SECTRA: sparse structured text rationalization,” in Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (Punta Cana: Association for Computational Linguistics), 6534–6550.

Gunning, D., Stefik, M., Choi, J., Miller, T., Stumpf, S., and Yang, G.-Z. (2019). Xai–explainable artificial intelligence. Sci. Robot . 4:37. doi: 10.1126/scirobotics.aay7120

Gurrapu, S., Huang, L., and Batarseh, F. A. (2022). “EXCLAIM: explainable neural claim verification using rationalization,” in 2022 IEEE 29th Annual Software Technology Conference (STC) (IEEE), 19–26.

Gurrapu, S., Kulkarni, A., Huang, L., Lourentzou, I., and Batarseh, F. A. (2023). Rationalization for explainable NLP: a survey. Front. Artif. Intell . 6:1225093. doi: 10.3389/frai.2023.1225093

Guyon, I., Cawley, G. C., Dror, G., and Lemaire, V. (2011). “Results of the active learning challenge,” in Active Learning and Experimental Design Workshop in Conjunction With AISTATS 2010. JMLR Workshop and Conference Proceedings (Sardinia), 19–45.

Hanselowski, A., Stab, C., Schulz, C., Li, Z., and Gurevych, I. (2019). “A richly annotated corpus for different tasks in automated fact-checking,” in Proceedings of the 23rd Conference on Computational Natural Language Learning (CoNLL) (Hong Kong: Association for Computational Linguistics), 493–503.

Hartmann, M., and Sonntag, D. (2022). “A survey on improving NLP models with human explanations,” in ACL Workshop on Learning with Natural Language Supervision (Dublin: Association for Computational Linguistics).

Hase, P., and Bansal, M. (2020). “Evaluating Explainable AI: Which Algorithmic Explanations Help Users Predict Model Behavior?” in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (Association for Computational Linguistics), 5540–5552.

Hase, P., and Bansal, M. (2022). “When can models learn from explanations? A formal framework for understanding the roles of explanation data,” in Proceedings of the First Workshop on Learning with Natural Language Supervision (Dublin: Association for Computational Linguistics), 29–39.

Hase, P., Zhang, S., Xie, H., and Bansal, M. (2020). “Leakage-adjusted simulatability: can models generate non-trivial explanations of their behavior in natural language?” in Findings of the Association for Computational Linguistics: EMNLP 2020 (Association for Computational Linguistics), 4351–4367.

Hayati, S. A., Kang, D., and Ungar, L. (2021). “Does bert learn as humans perceive? Understanding linguistic styles through lexica,” in Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (Punta Cana: Association for Computational Linguistics), 6323–6331.

He, P., Liu, X., Gao, J., and Chen, W. (2020). “DEBERTA: decoding-enhanced bert with disentangled attention,” in International Conference on Learning Representations (Vienna: ICLR).

Hochreiter, S., and Schmidhuber, J. (1997). Long short-term memory. Neural Comput . 9, 1735–1780.

Jacovi, A., and Goldberg, Y. (2021). Aligning faithful interpretations with their social attribution. Trans. Assoc. Comput. Linguist . 9, 294–310. doi: 10.48550/arXiv.2006.01067

Jain, S., and Wallace, B. C. (2019). “Attention is not explanation,” in Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) (Minneapolis, MN: Association for Computational Linguistics), 3543–3556.

Jain, S., Wiegreffe, S., Pinter, Y., and Wallace, B. C. (2020). “Learning to faithfully rationalize by construction,” in 58th Annual Meeting of the Association for Computational Linguistics, ACL 2020 (Association for Computational Linguistics), 4459–4473.

Jakobsen, T. S. T., Cabello, L., and Søgaard, A. (2023). “Being right for whose right reasons?” in The 61st Annual Meeting Of The Association For Computational Linguistics (Toronto, ON: Association for Computational Linguistics).

Jang, M., and Lukasiewicz, T. (2021). Are training resources insufficient? Predict first then explain! arXiv preprint arXiv:2110.02056 . doi: 10.48550/arXiv.2110.02056

Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., et al. (2023). Survey of hallucination in natural language generation. ACM Comput. Surv . 55, 1–38. doi: 10.1145/3571730

Jiang, H., Duan, J., Qu, Z., and Wang, J. (2023). You only forward once: prediction and rationalization in a single forward pass. arXiv preprint arXiv:2311.02344 . doi: 10.48550/arXiv.2311.02344

Jiang, X., Ye, H., Luo, Z., Chao, W., and Ma, W. (2018). “Interpretable rationale augmented charge prediction system,” in Proceedings of the 27th International Conference on Computational Linguistics: System Demonstrations (Santa Fe: Association for Computational Linguistics), 146–151.

Joshi, B., Liu, Z., Ramnath, S., Chan, A., Tong, Z., Nie, S., et al. (2023). “Are machine rationales (not) useful to humans? Measuring and improving human utility of free-text rationales,” in Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (Toronto, ON: Association for Computational Linguistics), 7103–7128.

Kanchinadam, T., Westpfahl, K., You, Q., and Fung, G. (2020). “Rationale-based human-in-the-loop via supervised attention,” in DaSH@ KDD (Association for Computing Machinery).

Kandul, S., Micheli, V., Beck, J., Kneer, M., Burri, T., Fleuret, F., et al. (2023). Explainable AI: A Review of the Empirical Literature . SSRN.

Kowsari, K., Jafari Meimandi, K., Heidarysafa, M., Mendu, S., Barnes, L., and Brown, D. (2019). Text classification algorithms: a survey. Information 10:150. doi: 10.48550/arXiv.1904.08067

Kumar, S., and Talukdar, P. (2020). “NILE : natural language inference with faithful natural language explanations,” in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics , 8730–8742.

Kung, P. N., Yang, T. H., Chen, Y. C., Yin, S. S., and Chen, Y. N. (2020). “Zero-shot rationalization by multi-task transfer learning from question answering,” in Findings of the Association for Computational Linguistics: EMNLP 2020 (Association for Computational Linguistics), 2187–2197.

Kunz, J., Jirenius, M., Holmström, O., and Kuhlmann, M. (2022). “Human ratings do not reflect downstream utility: a study of free-text explanations for model predictions,” in Proceedings of the Fifth BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP (Abu Dhabi: Association for Computational Linguistics), 164–177.

Lai, S., Xu, L., Liu, K., and Zhao, J. (2015). “Recurrent convolutional neural networks for text classification,” in Twenty-Ninth AAAI Conference on Artificial Intelligence (Burnaby, BC: PKP PS).

Lakhotia, K., Paranjape, B., Ghoshal, A., Yih, W. T., Mehdad, Y., and Iyer, S. (2021). “FID-EX: improving sequence-to-sequence models for extractive rationale generation,” in Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (Association for Computational Linguistics), 3712–3727.

Lan, Z., Chen, M., Goodman, S., Gimpel, K., Sharma, P., and Soricut, R. (2019). “ALBERT: a lite bert for self-supervised learning of language representations,” in International Conference on Learning Representations (Addis Ababa: ICLR).

Lei, T. (2017). Interpretable Neural Models for Natural Language Processing (Ph.D. thesis). Massachusetts Institute of Technology, Cambridge, MA, United States.

Lei, T., Barzilay, R., and Jaakkola, T. (2016). “Rationalizing neural predictions,” in Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing (Austin, TX: Association for Computational Linguistics), 107–117.

Lertvittayakumjorn, P., and Toni, F. (2019). “Human-grounded evaluations of explanation methods for text classification,” in Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) (Hong Kong: Association for Computational Linguistics), 5195–5205.

Lewis, M., Liu, Y., Goyal, N., Ghazvininejad, M., Mohamed, A., Levy, O., et al. (2020). “BART: denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension,” in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics , 7871–7880.

Li, D., Tao, J., Chen, Q., and Hu, B. (2021). You can do better! If you elaborate the reason when making prediction. arXiv preprint arXiv:2103.14919 . doi: 10.48550/arXiv.2103.14919

Liang, H., He, E., Zhao, Y., Jia, Z., and Li, H. (2022). Adversarial attack and defense: a survey. Electronics 11:1283. doi: 10.48550/arXiv.1810.00069

Liu, H., Yin, Q., and Wang, W. Y. (2019a). “Towards explainable NLP: a generative explanation framework for text classification,” in Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics , 5570–5581.

Liu, W., Wang, H., Wang, J., Li, R., Li, X., Zhang, Y., et al. (2023). “MGR: multi-generator based rationalization,” in Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (Toronto, ON: Association for Computational Linguistics), 12771–12787.

Liu, W., Wang, H., Wang, J., Li, R., Yue, C., and Zhang, Y. (2022). FR: folded rationalization with a unified encoder. Adv. Neural Inform. Process. Syst . 35, 6954–6966. doi: 10.48550/arXiv.2209.08285

Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., et al. (2019b). RoBERTa: a robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692 . doi: 10.48550/arXiv.1907.11692

Lundberg, S. M., and Lee, S. I. (2017). A unified approach to interpreting model predictions. Adv. Neural Inform. Process. Syst . 30:7874. doi: 10.48550/arXiv.1705.07874

Lyu, Q., Apidianaki, M., and Callison-Burch, C. (2024). Towards faithful model explanation in NLP: a survey. Comput. Linguist . 2024, 1–70. doi: 10.48550/arXiv.2209.11326

Lyu, Q., Zhang, H., Sulem, E., and Roth, D. (2021). “Zero-shot event extraction via transfer learning: challenges and insights,” in Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers) (Association for Computational Linguistics), 322–332.

Maas, A., Daly, R. E., Pham, P. T., Huang, D., Ng, A. Y., and Potts, C. (2011). “Learning word vectors for sentiment analysis,” in Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (Portland: Association for Computational Linguistics), 142–150.

Madani, M. R. G., and Minervini, P. (2023). “REFER: an end-to-end rationale extraction framework for explanation regularization,” in Proceedings of the 27th Conference on Computational Natural Language Learning (CoNLL) (Singapore: Association for Computational Linguistics), 587–602.

Madsen, A., Reddy, S., and Chandar, S. (2022). Post-hoc interpretability for neural NLP: a survey. ACM Comput. Surv . 55, 1–42. doi: 10.1145/3546577

Mahoney, C., Gronvall, P., Huber-Fliflet, N., and Zhang, J. (2022). “Explainable text classification techniques in legal document review: locating rationales without using human annotated training text snippets,” in 2022 IEEE International Conference on Big Data (Big Data) (Osaka: IEEE), 2044–2051.

Majumder, B. P., Camburu, O., Lukasiewicz, T., and Mcauley, J. (2022). “Knowledge-grounded self-rationalization via extractive and natural language explanations,” in International Conference on Machine Learning (Baltimore, MD: PMLR), 14786–14801.

Marasović, A., Beltagy, I., Downey, D., and Peters, M. E. (2022). “Few-shot self-rationalization with natural language prompts,” in Findings of the Association for Computational Linguistics: NAACL 2022 (Seattle, WA: Association for Computational Linguistics), 410–424.

Marasović, A., Bhagavatula, C., Park, J. S., Bras, R. L., Smith, N. A., and Choi, Y. (2020). “Natural Language Rationales With Full-Stack Visual Reasoning: From Pixels to Semantic Frames to Commonsense Graphs,” in Findings of the Association for Computational Linguistics: EMNLP 2020 (Association for Computational Linguistics), 2810–2829.

Mathew, B., Saha, P., Yimam, S. M., Biemann, C., Goyal, P., and Mukherjee, A. (2021). HateXplain: a benchmark dataset for explainable hate speech detection. arXiv . 35, 14867–14875. doi: 10.48550/arXiv.2012.10289

McAuley, J., Leskovec, J., and Jurafsky, D. (2012). “Learning attitudes and attributes from multi-aspect reviews,” in 2012 IEEE 12th International Conference on Data Mining (Brussels: IEEE), 1020–1025.

Mendez, E., Schlegel, V., and Batista-Navarro, R. (2022). “RaFoLa: a rationale-annotated corpus for detecting indicators of forced labour,” in Proceedings of the Thirteenth Language Resources and Evaluation Conference , eds. N. Calzolari, F. Béchet, P. Blache, K. Choukri, C. Cieri, T. Declerck, et al. (Marseille: European Language Resources Association), 3610–3625.

Meng, Y., Huang, J., Zhang, Y., and Han, J. (2022). Generating training data with language models: towards zero-shot language understanding. Adv. Neural Inform. Process. Syst . 35, 462–477. doi: 10.48550/arXiv.2202.04538

Miller, T. (2019). Explanation in artificial intelligence: insights from the social sciences. Artif. Intell . 267, 1–38. doi: 10.1016/j.artint.2018.07.007

Minervini, P., Franceschi, L., and Niepert, M. (2023). “Adaptive perturbation-based gradient estimation for discrete latent variable models,” in Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 37 (Burnaby, BC: PKP PS), 9200–9208.

Mohseni, S., Block, J. E., and Ragan, E. D. (2018). A human-grounded evaluation benchmark for local explanations of machine learning. arXiv preprint arXiv:1801.05075 . doi: 10.48550/arXiv.1801.05075

Mukherjee, S. (2019). General information bottleneck objectives and their applications to machine learning. arXiv preprint arXiv:1912.06248 . doi: 10.48550/arXiv.1912.06248

Narang, S., Raffel, C., Lee, K., Roberts, A., Fiedel, N., and Malkan, K. (2020). WT5?! training text-to-text models to explain their predictions. arXiv preprint arXiv:2004.14546 . doi: 10.48550/arXiv.2004.14546

Niculae, V., and Martins, A. (2020). LP-SparseMAP: differentiable relaxed optimization for sparse structured prediction. In International Conference on Machine Learning (Baltimore, MD: PMLR), 7348–7359.

Niu, Z., Zhong, G., and Yu, H. (2021). A review on the attention mechanism of deep learning. Neurocomputing 452, 48–62. doi: 10.1016/j.neucom.2021.03.091

Otter, D. W., Medina, J. R., and Kalita, J. K. (2020). A survey of the usages of deep learning for natural language processing. IEEE Trans. Neural Netw. Learn. Syst . 32, 604–624. doi: 10.1109/TNNLS.2020.2979670

Palmer, J., Kreutz-Delgado, K., Rao, B., and Wipf, D. (2005). Variational em algorithms for non-gaussian latent variable models. Adv. Neural Inform. Process. Syst . 2005:18.

Paranjape, B., Joshi, M., Thickstun, J., Hajishirzi, H., and Zettlemoyer, L. (2020). “An information bottleneck approach for controlling conciseness in rationale extraction,” in Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) (Association for Computational Linguistics).

Park, D. H., Hendricks, L. A., Akata, Z., Rohrbach, A., Schiele, B., Darrell, T., et al. (2018). “Multimodal explanations: Justifying decisions and pointing to the evidence,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (Salt Lake City, UT: IEEE), 8779–8788.

Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., Sutskever, I., et al. (2019). Language models are unsupervised multitask learners. OpenAI Blog 1:9.

Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., et al. (2020). Exploring the limits of transfer learning with a unified text-to-text transformer. J. Mach. Learn. Res . 21, 1–67. doi: 10.48550/arXiv.1910.10683

Rajani, N. F., McCann, B., Xiong, C., and Socher, R. (2019). “Explain yourself! leveraging language models for commonsense reasoning,” in Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (Florence: Association for Computational Linguistics), 4932–4942.

Ribeiro, M. T., Singh, S., and Guestrin, C. (2016a). “Why should i trust you?” explaining the predictions of any classifier,” in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (New York, NY: Association for Computing Machinery), 1135–1144.

Ribeiro, M. T., Singh, S., and Guestrin, C. (2016b). Model-agnostic interpretability of machine learning. arXiv preprint arXiv:1606.05386 . doi: 10.48550/arXiv.1606.05386

Ross, A., Peters, M. E., and Marasović, A. (2022). “Does self-rationalization improve robustness to spurious correlations?” in Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing (Florence: Association for Computational Linguistics), 7403–7416.

Sen, C., Hartvigsen, T., Yin, B., Kong, X., and Rundensteiner, E. (2020). “Human attention maps for text classification: do humans and neural networks focus on the same words?” in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (Association for Computational Linguistics), 4596–4608.

Simonyan, K., Vedaldi, A., and Zisserman, A. (2014). “Deep inside convolutional networks: visualising image classification models and saliency maps,” in Proceedings of the International Conference on Learning Representations (ICLR) (Banff, AB: ICLR).

Socher, R., Perelygin, A., Wu, J., Chuang, J., Manning, C. D., Ng, A., et al. (2013). “Recursive deep models for semantic compositionality over a sentiment treebank,” in Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing (Seattle, WA: Association for Computational Linguistics), 1631–1642.

Strout, J., Zhang, Y., and Mooney, R. J. (2019). “Do human rationales improve machine explanations?” in Proceedings of the 2019 ACL Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP (Florence: Association for Computational Linguistics), 56–62.

Sundararajan, M., Taly, A., and Yan, Q. (2017). “Axiomatic attribution for deep networks,” in International Conference on Machine Learning (Sydney, NSW: PMLR), 3319–3328.

Sutskever, I., Vinyals, O., and Le, Q. V. (2014). Sequence to sequence learning with neural networks. Adv. Neural Inform. Process. Syst . 27:3215. doi: 10.48550/arXiv.1409.3215

Tan, C. (2022). “On the diversity and limits of human explanations,” in Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Seattle, WA: Association for Computational Linguistics).

Thorne, J., Vlachos, A., Christodoulopoulos, C., and Mittal, A. (2018). “Fever: a large-scale dataset for fact extraction and verification,” in Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers) (New Orleans, LA: Association for Computational Linguistics), 809–819.

Tjoa, E., and Guan, C. (2020). A survey on explainable artificial intelligence (XAI): toward medical XAI. IEEE Trans. Neural Netw. Learn. Syst . 32, 4793–4813. doi: 10.48550/arXiv.1907.07374

Tornqvist, M., Mahamud, M., Guzman, E. M., and Farazouli, A. (2023). “EXASAG: explainable framework for automatic short answer grading,” in Proceedings of the 18th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2023) (Toronto, ON: Association for Computational Linguistics), 361–371.

Vashishth, S., Upadhyay, S., Tomar, G. S., and Faruqui, M. (2019). Attention interpretability across NLP tasks. arXiv preprint arXiv:1909.11218 . doi: 10.48550/arXiv.1909.11218

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., et al. (2017). Attention is all you need. Adv. Neural Inform. Process. Syst . 30:3762. doi: 10.48550/arXiv.1706.03762

Vijayan, V. K., Bindu, K., and Parameswaran, L. (2017). “A comprehensive study of text classification algorithms,” in 2017 International Conference on Advances in Computing, Communications and Informatics (ICACCI) (Udupi: IEEE), 1109–1113.

Wang, H., and Dou, Y. (2022). “Recent development on extractive rationale for model interpretability: a survey,” in 2022 International Conference on Cloud Computing, Big Data and Internet of Things (3CBIT) (Wuhan: IEEE), 354–358.

Wang, H., Lu, Y., and Zhai, C. (2010). “Latent aspect rating analysis on review text data: a rating regression approach,” in Proceedings of the 16th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , 783–792.

Wei, J., Bosma, M., Zhao, V. Y., Guu, K., Yu, A. W., Lester, B., et al. (2021). “Finetuned language models are zero-shot learners,” in International Conference on Learning Representations (ICLR).

Wiegreffe, S., and Marasović, A. (2021). “Teach me to explain: a review of datasets for explainable natural language processing,” in Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 1) (NeurIPS).

Wiegreffe, S., Marasović, A., and Smith, N. A. (2021). “Measuring association between labels and free-text rationales,” in Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (Punta Cana: Association for Computational Linguistics), 10266–10284.

Wiegreffe, S., and Pinter, Y. (2019). “Attention is not explanation,” in Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) (Hong Kong: Association for Computational Linguistics), 11–20.

Williams, R. J. (1992). Simple statistical gradient-following algorithms for connectionist reinforcement learning. Machine learn . 8, 229–256.

Xue, L., Constant, N., Roberts, A., Kale, M., Al-Rfou, R., Siddhant, A., et al. (2021). “MT5: a massively multilingual pre-trained text-to-text transformer,” in Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Association for Computational Linguistics), 483–498.

Yao, B., Sen, P., Popa, L., Hendler, J., and Wang, D. (2023). “Are human explanations always helpful? Towards objective evaluation of human natural language explanations,” in Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (Toronto, ON: Association for Computational Linguistics), 14698–14713.

Yu, M., Chang, S., Zhang, Y., and Jaakkola, T. S. (2019). “Rethinking cooperative rationalization: introspective extraction and complement control,” in Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) , 4094–4103.

Yuan, J., Sun, S., Omeiza, D., Zhao, B., Newman, P., Kunze, L., et al. (2024). RAG-Driver: generalisable driving explanations with retrieval-augmented in-context learning in multi-modal large language model. arXiv preprint arXiv:2402.10828 . doi: 10.48550/arXiv.2402.10828

Zaidan, O., Eisner, J., and Piatko, C. (2007). “Using “annotator rationales” to improve machine learning for text categorization,” in Human Language Technologies 2007: The Conference of the North American Chapter of the Association for Computational Linguistics; Proceedings of the Main Conference (Rochester, NY: Association for Computational Linguistics), 260–267.

Zeiler, M. D., and Fergus, R. (2014). “Visualizing and understanding convolutional networks,” in European Conference on Computer Vision (Berlin: Springer), 818–833.

Zhang, D., Sen, C., Thadajarassiri, J., Hartvigsen, T., Kong, X., and Rundensteiner, E. (2021a). “Human-like explanation for text classification with limited attention supervision,” in 2021 IEEE International Conference on Big Data (Orlando, FL: IEEE), 957–967.

Zhang, J., Kim, J., O'Donoghue, B., and Boyd, S. (2021b). “Sample efficient reinforcement learning with reinforce,” in Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 35 , 10887–10895.

Zhang, J., Zheng, L., Guo, D., and Wang, M. (2024). Training a small emotional vision language model for visual art comprehension. arXiv preprint arXiv:2403.11150 . doi: 10.48550/arXiv.2403.11150

Zhang, W. E., Sheng, Q. Z., Alhazmi, A., and Li, C. (2020). Adversarial attacks on deep-learning models in natural language processing: a survey. ACM Trans. Intell. Syst. Technol . 11, 1–41. doi: 10.1145/3374217

Zhang, Y., Marshall, I., and Wallace, B. C. (2016). “Rationale-augmented convolutional neural networks for text classification,” in Proceedings of the Conference on Empirical Methods in Natural Language Processing. Conference on Empirical Methods in Natural Language Processing, Vol. 2016 (NIH Public Access), 795.

Zhao, H., Chen, H., Yang, F., Liu, N., Deng, H., Cai, H., et al. (2023). Explainability for large language models: a survey. ACM Trans. Intell. Syst. Technol . 2023:1029. doi: 10.48550/arXiv.2309.01029

Zhao, X., and Vydiswaran, V. V. (2021). “LIREX: augmenting language inference with relevant explanations,” in Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 35 , 14532–14539.

Zhou, W., Hu, J., Zhang, H., Liang, X., Sun, M., Xiong, C., et al. (2020). Towards interpretable natural language understanding with explanations as latent variables. Adv. Neural Inform. Process. Syst . 33, 6803–6814. doi: 10.48550/arXiv.2011.05268

Zhou, Y., Zhang, Y., and Tan, C. (2023). “FLAME: few-shot learning from natural language explanations,” in Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (Toronto, ON: Association for Computational Linguistics), 6743–6763.

Zini, J. E., and Awad, M. (2022). On the explainability of natural language processing deep models. ACM Comput. Surv . 55, 1–31. doi: 10.1145/3529755

Performance of rationalization approaches

Table A1 presents the breakdown results for rationalization approaches according to what has been reported for each author on the MovieReviews ( Zaidan et al., 2007 ), SST ( Socher et al., 2013 ), and the FEVER ( Thorne et al., 2018 ) datasets. The predictive performance is evaluated using the F1 Score (F1), and the quality of the produced rationales is assessed using Sufficiency (Suff) and Comprehensiveness (Comp).

Table A1 . Performance of different rationalization approaches on the MovieReviews, SST, and FEVER datasets.

Keywords: Natural Language Processing, text classification, Explainable Artificial Intelligence, rationalization, language explanations

Citation: Mendez Guzman E, Schlegel V and Batista-Navarro R (2024) From outputs to insights: a survey of rationalization approaches for explainable text classification. Front. Artif. Intell. 7:1363531. doi: 10.3389/frai.2024.1363531

Received: 30 December 2023; Accepted: 02 July 2024; Published: 23 July 2024.

Reviewed by:

Copyright © 2024 Mendez Guzman, Schlegel and Batista-Navarro. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

*Correspondence: Erick Mendez Guzman, erick.mendezguzman@manchester.ac.uk

Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.

We use our proprietary methodology to produce these rankings. Learn more →

The Top Artificial Intelligence Colleges in Texas

Share on Facebook
Email this Page
Share on LinkedIn

The Top Artificial Intelligence Colleges in Texas
Rank	School	Location
1		Dallas, Texas
2		Austin, Texas
3		Denton, Texas
4		San Antonio, Texas
5		College Station, Texas
6		Commerce, Texas
7		Richardson, Texas

We have researched the top colleges for Artificial Intelligence degrees in Texas.

Our ranking is designed for prospective students, parents, teachers, and guidance counselors to be informed of college and degree options.

Undergraduate and graduate students can pursue degrees in Artificial Intelligence, including bachelor's, master's, and a PhD (doctorate) degree.

Texas is a technological hub in many industries, including the Artificial Intelligence and Machine Learning industries. The Institute for Foundations of Machine Learning (IFML) is located in Texas.

Designated by the National Science Foundation (NSF) in 2020, IFML develops the key foundational tools for the next decade of AI innovation. Our institute comprises researchers from The University of Texas at Austin, University of Washington, Wichita State University, and Microsoft Research. IFML

Southern Methodist University

Lyle School of Engineering

Dallas, Texas

Southern Methodist University (SMU), in Dallas, Texas, offers the online Master of Science in Computer Science with an Artificial Intelligence Specialization.

This degree teaches data mining, machine learning, and cloud computing. GRE waivers may be available to qualifying students.

This program's resources are available to online students 24 hours a day, 7 days a week. Online students can participate in weekly sessions face-to-face online with professors. There is effectively no difference with on-campus programs with respect to instructor access.

Credits: 30 total

Courses include:

Artificial Intelligence
Machine Learning in Python
Computer Architecture
Algorithm Engineering

This program has multiple start dates throughout the year. Applicants will need a Bachelor of Science in Computer Engineering, Computer Science, or a related engineering subject.

SMU also offers a Master of Science in Data Science with a Machine Learning specialization.

This program teaches students how to use machine learning to make intelligent decisions and interpretations using big data sets.

The University of Texas at Austin

College of Natural Sciences

Austin, Texas

The University of Texas at Austin offers an online Master of Artificial Intelligence, which also includes instruction in Machine Learning.

Online students learn the following AI subjects:

ethics in AI
case studies in machine learning
reasoning under uncertainty

The Ethics in AI class is the only required course, and the remaining 27 hours are Electives.

Elective courses include:

Case studies in Machine Learning
Online Learning and Optimization
AI in Healthcare
Natural Language Processing
Automated Logical Reasoning

Credits: 30 credit hours (each course is 3 hours, so there are 10 courses)

Classes are asynchronous, which means that there is no set time to attend classes, which allows for flexibility.

University of North Texas

College of Engineering

Denton, Texas

University of North Texas (UNT) offers the Master of Science in Artificial Intelligence graduate program.

According to UNT, this program is the only standalone master's degree in artificial intelligence in the state of Texas.

Students can specialize in:

autonomous systems
biomedical engineering
machine learning

Credits: 33 hours (bridge and core courses: 18 hours, validation and testing courses: 3 hours, and concentration courses: 12 hours).

Bridge courses teach relevant programming skills.

Core courses teaches competencies in these subjects:

feature engineering
big data and data science
deep learning

Concentration courses cover areas such as autonomous systems and machine learning.

The University of Texas at San Antonio

School of Data Science

San Antonio, Texas

University of Texas at San Antonio offers a Bachelor of Science Artificial Intelligence: Multidisciplinary Studies degree.

This bachelor's teaches students AI through examining and understanding human intelligence processes, self-correcting, and reasoning.

This degree is interdisciplinary, with students studying AI within these domains:

information systems
mathematics
electrical and computer engineering
computer science

Students can customize their program or choose from a pre-established program around a particular niche. The AI program is one of the niche options that students may choose.

Artificial Intelligence – Multidisciplinary Studies (BS)

Texas A & M University-College Station

College Station, Texas

Texas A&M University offers a Ph.D. in Civil Engineering with an Artificial Intelligence and Data Science track. This PhD degree is interdisciplinary.

Areas of research include:

automated infrastructure monitoring
disaster and multi-hazard engineering informatics
transportation big data
city-scale predictive analytics
computer-aided infrastructure engineering and management
sustainable decision making
data-driven approaches to understand environmental processes
environmental systems analytics
urban intelligence and computing

Credits: 64

Spatial Statistics
Machine Learning with Networks
Data Mining and Analysis
Statistics in Research I

Texas A & M University-Commerce

Commerce, Texas

Texas A&M University offers a campus-based, hybrid or online Master of Science in Artificial Intelligence. The hybrid option allows students to take some classes online, and some on-campus.

Credits: 34 - 37

There are non-thesis and thesis options available.

Students can choose from four emphases:

Computational Linguistics
Computer Science
Mathematics
Foundations of AI
Python Programming for AI
Machine Learning for AI
Image Analysis and Recognition with Learning
Seminar in AI Ethics
Introduction to Learning Technology

Artificial intelligence (MS)

The University of Texas at Dallas

Department of Computer Science

Richardson, Texas

University of Texas at Dallas offers a Bachelor of Science in Computer Science with applications in Artificial Intelligence, Machine Learning, Virtual Reality, and other subjects.

Introduction to Human-Computer Interaction
Data and Applications Security
Automata Theory
Compiler Design
Introduction to Machine Learning
Object-Oriented Design

Automata Theory deals with theories about abstract machines answering questions; which includes positing the kinds of questions that can be answered and the kinds that cannot. Abstract machines are models that show how machine functions answer questions.

An example of an abstract machine is the Turing machine, which follows a set of instructions that determines its tasks.

More resources:

The Best Artificial Intelligence Colleges
The Best Online Bachelor's in Artificial Intelligence
The Best Online Artificial Intelligence Degrees
The Best Online Data Analytics Colleges
The Best Online Machine Learning Master's Degrees

Academia.edu no longer supports Internet Explorer.

To browse Academia.edu and the wider internet faster and more securely, please take a few seconds to upgrade your browser .

Enter the email address you signed up with and we'll email you a reset link.

We're Hiring!
Help Center

Application of Artificial Intelligence in Architectural Design [Ph.D. Dissertation]

Prior knowledge plays a major role in architectural design. This knowledge pertains to the products and processes of design. Utilizing computers as a design medium requires the representation of such knowledge for reasoning purposes. The choice of what to represent from these concepts (i.e., products and processes) is critical in the utilization of knowledge-based systems in design. The goals of this research are the enhancement of architectural flexibility and generative capabilities in design environments. Both goals are largely influenced by the knowledge represented in design environments. Architectural flexibility pertains to the compositional diversity required when using this stored knowledge to address the design of different space and building types. Generative capabilities pertain to the application of design processes to propose possible solutions, that add to the explication of the problems we are facing. The problems of this research pertain to the level and type of decomposition that is applied on the concepts of architectural design, representation of these concepts, and their utilization in knowledge-based design systems. To enhance the flexibility and generation capabilities of design environments, the research proposes to aggregate space functions to the set of human activities that will be performed in them. This functional decomposition provides the basis for refining, adapting, and creating new space types from existing knowledge about human activities. Consequently, building types can be refined, adapted, and created from their functional aggregates (i.e., spaces). The contributions of this research are based on the ability to represent, manipulate, and create space functions. Consequently, it is possible to describe and manipulate different building types. which achieve the goals of this research. Those contributions are theoretically grounded on cognitive and AI-based design research, and technically examined through the design and implementation of a knowledge-based experimental design environment (APE-1), that is intended to support architects in an early stage of design (i.e., architectural programming).

Related Papers

M. Sherif El-Attar

This paper focuses on the 'type' of design knowledge used in computational design environments. The paper discuss prototypical information in knowledge-based design systems, and suggests a functional type of knowledge for usage in such systems in order to achieve a degree of generalization in the domain of space design. A function in an architectural space depends on the activities that accomplish it. The paper focuses on human activities and how they may provide the means to semantically describe architectural spaces in different building types.

Proceedings of S.Arch 2020, the 7th international conference on architecture and built environment

Giuseppe Gallo , fulvio wirz

The proliferation of data together with the increase of computing power in the last decade has triggered a new interest in artificial intelligence methods. Machine learning and in particular deep learning techniques, inspired by the topological structure of neurons network in brains, are omnipresent in the IT discourse, and generated new enthusiasms and fears in our society. These methods have already shown great effectiveness in fields far from architecture and have long been exploited in software that we use every day. Many computing libraries are available for anyone with some programming skills and allow them to "train" a neural network based on several types of data. The world of architecture has not remained external to this phenomenon: many researchers are working on the applications of artificial intelligence to architectural design, a few design software allow exploiting machine learning algorithms, and some large architectural firms have begun to experiment with deep learning methods to put into practice data accumulated over years of profession, with a special interest in environmental sustainability and building performance. If on the one hand, these techniques promise great results, on the other we are still in an exploratory phase. It is then necessary, in our opinion, to understand what the roles of this technology could be within the architectural design process, and with which scopes they can facilitate such a complex profession as that of the architect. On this subject we made ten interviews with as many designers and researchers in the AEC industry, In the article we will report a summary of their testimonies, comparing and commenting on the responses of the designers, with the aim of understanding the potentials of using artificial intelligence methods within the design process, report their perceptions on how artificial intelligence techniques can affect the architect's approach to the project, concluding with some reflections on the critical issues identified during the interviews with the designers.

Computer-Aided Design

Yehuda E. Kalay

… of eCAADe 01 …

Gianfranco Carrara , Antonio Fioravanti

Prof. Rabee M. Reffat

Silvia Gargaro , Antonio Fioravanti

From the earliest stages of the Architectural Design Process, designers have to take a lot of design decisions mostly based on “Context”. The present research is aimed at developing a Context Knowledge Model to improve the representation of ‘Context’ for architectural design. ‘Context’ has been analysed and formalized by means of Ontologies related to the entities most frequently involved in architectural design, namely environmental, social, economic and normative entities. The development of such a model to manage ‘Context’ parameters can improve the knowledge of ‘Context’ of designers involved in a design project in order to advise them of how it affects their design solutions. Moreover, Artificial Intelligence techniques have been explored to improve its performance.

Proceedings of Expert …

Alasdair Turner

Gianfranco Carrara

The 54th International Conference of the Architectural Science Association

Manuel Mühlbauer

This position paper describes a pathway and methodology towards creative systems in architectural design. Drawing from creativity research and strategic design methods, an agile approach to exploration of deep learning technology in the context of architectural optimisation was developed. The investigation proposed and defined the nature of a framework, which explored ways of integrating architectural shape design with machine intelligence. Furthermore, the paper elaborates the implications and potential for impact of deep learning techniques on advancing human-computer-interaction for architectural optimisation. However, the described framework might be used as a design scheme for an active tool to drive design processes and support decision-making in early stages of architectural design. The components of the framework defined interfaces and critical points of investigation for application of the presented methodology in creative practice. In this way the research contributes to the theoretical and methodological development of creative systems research. At its heart, this generative design study involved the definition of a clear research trajectory, challenges and opportunities of supporting creative practice by means of design systems. Finally, the potential of machine intelligence to generate creative work with and without human guidance or performance criteria was examined.

Digital Design, proceedings of eCAADe

RELATED PAPERS

Sven Bertel

Journal of Computational Design

Kamile Öztürk Kösenciğ

Alessio Erioli

Nikos Salingaros

Harvard University

Stanislas Chaillou

Ciric, D. 2022. Generative design principles and algorithmic thinking in architectural plan variations: human to machine approach. In Soddu, C. & Colabella, E. (Eds.), XXV Generative Art 2022: proceedings of the XXV GA conference (pp. 144-156). Rome: Domus Argenia Publishing,

Dragana Ciric

Eduardo Lyon

Architectus

Anna Jaruga-Rozdolska

eCAADe proceedings

Kacper Radziszewski

Iniciativa Digital Politècnica. Oficina de Publicacions Acadèmiques Digitals de la UPC eBooks

Constantin Victor Spiridonidis

Derya Güleç Özer

Blucher Design Proceedings

Christoph Langenhan

Proceedings of the 14th …

Pieter Pauwels

Architectural Intelligence

George Vlachodimos

Nizam Onur Sönmez

MSA Engineering Journal

ahmed saleh

Voiland College Welcomes New Faculty for 2024-25

Meet the cohort of new faculty members joining Voiland College during the 2024-2025 school year.

Hala Barakat

Assistant Professor, Architecture, School of Design and Construction.

Hala Barakat focuses her studies on early design pedagogy and social justice in architecture. She holds a Master of Architecture and a Master of Urban and Community design from the University of South Florida’s School of Architecture and Community Design. Before coming to WSU, she was an assistant professor at the University of Idaho.

Assistant Professor, School of Electrical Engineering and Computer Science.

Dr. Cao joins WSU from the Hong Kong University of Science and Technology, where he has been an assistant professor with the Department of Electronic and Computer Engineering since 2021. He also was a postdoctoral research associate at Princeton University and the Coordinated Science Lab at the University of Illinois, Urbana-Champaign. His research entails developing and analyzing distributed/online learning and optimization algorithms with various practical considerations, including communication efficiency, imperfect feedback information, imperfect information transmission, performativity, and data privacy & security. He is a senior member of IEEE, an editor for the IEEE Transactions on Wireless Communications and IEEE Transactions on Vehicular Technology, and the lead guest editor for the special issue on “Communication-Efficient Distributed Learning over Networks” in IEEE Journal on Selected Areas in Communications in 2023. He is a TPC member for MobiHoc 2022-2024, INFOCOM 2025, and WiOpt 2024. Dr. Cao holds a B.S. degree in electrical engineering from Shanghai Jiao Tong University, and a master’s and Ph.D. from the University of Maryland, College Park.

Neil Corrigan

Visiting Teaching Faculty, Cybersecurity.

Dr. Corrigan will serve as a teaching professor, instructing diverse cybersecurity courses. His career at Hanford in computer science spanned decades, concentrating on software development and cybersecurity. Alongside his industry tenure, he significantly contributed to WSU Tri-Cities as an adjunct faculty member in Computer Science. He has seamlessly integrated his practical software development experience into academia, imparting courses on software engineering principles and recently overseeing the senior software design project. His academic credentials include a PhD and master’s degree from Washington State University, complemented by a bachelor’s degree from the University of Utah.

Assistant Professor, School of Design and Construction.

Dr. Cronan is a landscape architect who will be focusing his work on food and the built environment. Before joining WSU, he was an assistant professor in landscape architecture and a senior research associate in the Center for Resilient Communities at the University of Idaho. He also was at the State University of New York College of Environmental Science and Forestry. He holds a PhD in Environmental Science from the University of Idaho, a master’s degree in landscape architecture from the University of Oregon, and a bachelor’s degree from Louisiana State University.

Seyyedmilad (Milad) Ebrahimi

Dr. Ebrahimi’s main areas of research interests include modeling and control of renewable energy systems, power systems, power electronics, and electrical machines; as well as power-hardware-in-the-loop real-time electromagnetic transient simulations, for the development of the next-generation sustainable energy systems. Before joining WSU, he was a postdoctoral research and teaching fellow at The University of British Columbia (UBC), Vancouver, Canada. He received his Ph.D. degree from UBC, and his M.Sc. and B.Sc. degrees from Sharif University of Technology, Iran, all in electrical engineering.

Mohamed Elmahallawy

Assistant Professor, Cybersecurity.

Dr. Elmahallawy conducted research on federated learning for LEO satellite constellations and medical applications, trustworthy artificial intelligence, and IoT. His research also encompasses cybersecurity in machine/federated learning. Prior to joining Washington State University Tri-Cities, he served as a postdoctoral fellow in the Department of Computer Science at Missouri University of Science and Technology. There, he focused on developing a global ML approach for enhancing miners’ safety in underground mines. He holds a PhD from Missouri University of Science and Technology, as well as master’s degrees from the University of Rostock in Germany and Benha University in Egypt. He completed his bachelor’s degree in Electronics and Communications Engineering at the Higher Institute of Engineering in Egypt.

Bill Furman

Instructor, Interior Design, School of Design and Construction.

Bill Furman aims to bridge the gap between academic experiences and disciplinary professional values, balancing imparting knowledge and skill, and preparing students for human-centered thinking that supports all end-users while simultaneously preparing them for careers as interior design practitioners.

Thomas Gilray

Associate Professor, School of Electrical Engineering and Computer Science.

Before joining WSU, Dr. Gilray was Victor Basili Fellow at the University of Maryland, at College Park, and then Assistant Professor in Computer Science at the University of Alabama at Birmingham. His research interests center around the design and scalable implementation of high-level programming languages and systems for reasoning automatically about programs. He has contributed to the design of tunable whole-program control-flow analyses, information-flow analyses, contract verification systems, and has invented novel Datalog-based languages for implementing these analyses efficiently—and declaratively. He has also contributed to high-performance-computing techniques for accelerating data-parallel relational algebra and sparse linear algebra on supercomputers. He holds a PhD from the University of Utah.

Rahul Gupta

Dr. Gupta’s research specializes in grid-aware control of distributed energy resources in active distribution networks, addressing various uncertainties through measurement-based and data-driven control and estimation schemes. Before joining Washington State University, he was a postdoctoral researcher at Georgia Tech’s School of Electrical and Computer Engineering, funded by the Swiss National Science Foundation (SNSF). At Georgia Tech, he developed methods for generating synthetic power distribution networks, hosting capacity analysis, and different fairness-aware operation and planning schemes. Rahul earned his M.Sc. and Ph.D. degrees in electrical engineering from the Swiss Federal Institute of Technology, Lausanne (EPFL) in 2018 and 2023, respectively. He received the EPFL PhD Thesis Distinction in Electrical Engineering in 2023 for his exceptional doctoral research.

Hyeyoung Koh

Assistant Professor, Department of Civil and Environmental Engineering.

Dr. Koh is conducting research in the area of intelligent and AI-aided design of steel and hybrid structures with an emphasis on their resilience and sustainability. Her group conducts research on the next generation of structural design and analysis by employing various methodological approaches including data analytics, distributed sensing, physical experiments, and computational modeling. Before joining WSU, she was a postdoctoral researcher at the University of Wisconsin, Madison. She received the 2024 Yoon Duk Kim Young Researcher Award from the Structural Stability Research Council (SSRC) within the American Institute of Steel Construction. She serves as vice-chair of the SSRC Task Group 03 – Steel Systems and as a committee member on the Stability Committee in the ASCE Engineering Mechanics Institute. Dr. Koh holds a PhD from University of Wisconsin, Madison, and a bachelor’s and master’s degree from Hanyang University in South Korea.

Parteek Kumar

Associate Professor Career Track, School of Electrical Engineering and Computer Science.

Dr. Kumar conducts research in machine learning, natural language processing, large language models, and AI applications for social good. Before joining WSU, he was a professor at Thapar Institute of Engineering and Technology, Patiala, India, and a visiting professor at Whitman College, USA, and the LAMBDA Lab at Tel Aviv University, Israel. He holds a doctorate from Thapar Institute, and a master’s from BITS Pilani, India. Dr. Kumar is an author of databases and data mining textbooks, including “Data Mining and Data Warehousing” published by Cambridge University Press. He leads significant research projects, teaches online courses on Udemy, and runs a YouTube channel.

Assistant Professor, Computer Science.

Dr. Li specializes in research on task and motion planning algorithms in robotics, optimization, parallel algorithms, and human-robot interaction (HRI). Prior to joining Washington State University Tri-Cities, she served as a graduate research assistant at the Dynamic Automata Lab (DyALab) at Colorado School of Mines, focusing on feasibility analysis in robot motion planning. She holds a PhD from Colorado School of Mines, master’s degrees from Worcester Polytechnic Institute and Rensselaer Polytechnic Institute, and a bachelor’s degree in mechanical engineering from the University of Science and Technology in China.

Assistant Professor, Electrical Engineering.

Dr. Liu conducted research on cyber-physical security of power systems, machine learning applications, and state estimation in smart grids. His work aims to advance smart grid cyber-physical security through innovative contributions to planning, operational strategies, and control methodologies. Leveraging integrated technologies such as computer and network systems, power system analysis, graph theory, game theory, and mathematical optimization, he seeks to transform the field. Prior to joining WSU Tri-Cities, he served as a Research Assistant Professor and as a Postdoc at Kansas State University. His focus there was on modeling, enhancing, and developing moving target defense (MTD) to detect and prevent cyberattacks against smart grids. In the summer of 2018, he was a graduate research intern at the National Renewable Energy Laboratory (NREL). He holds a PhD from Kansas State University, as well as along with master’s and bachelor’s degrees from Harbin Institute of Technology in China.

Nishtha Madaan

Dr. Madaan conducts research on large language model (LLM) safety, focusing on developing algorithms to enhance trust in LLMs. She has been a research staff member at IBM Research, where she has been an integral part of the team for nine years. Her academic background includes a PhD from the Indian Institute of Technology Delhi (IITD). She earned her B.Tech. and an MS from the International Institute of Information Technology Hyderabad, (IIITH). She has published various papers at renowned AI conferences and was recognized among the Top 100 Women in AI Ethics in 2019, being the only selection from India. She has mentored over 70 students throughout her career.

Satyajit Mojumder

Assistant Professor, School of Mechanical and Materials Engineering.

Dr. Mojumder’s research focuses on developing innovative computational methods to address large-scale challenges in advanced materials and manufacturing systems. He is exploring the integration of data science algorithms into computational methods, creating mechanistic reduced-order models crucial for tackling complex computational problems within the domain of multiscale, multifunctional materials produced through advanced manufacturing processes. These methods have led to the establishment of a startup company (HIDENN-AI, LLC) and secured funding for multiple SBIR/STTR projects, including NSF ACCESS computational resources for which he served as the Principal Investigator. He holds a bachelor and master’s degrees in mechanical engineering from Bangladesh University of Engineering and Technology, Dhaka, and a Ph.D. from Northwestern University. Before coming to WSU, he was a faculty member at Bangladesh University of Engineering and Technology. His collaborative research has resulted in 39 journal articles, 17 conference proceedings, two invited talks, four patents and disclosures, and fellowship grants and awards, including the Walter Murphy Fellowship, the Predictive Science and Engineering Design Fellowship, an NSF travel grant and the Northwestern TGS conference grant.

Wheaton Schroeder

Assistant Professor, Gene and Linda School of Chemical Engineering and Bioengineering.

Dr. Schroeder’s research is in the area of multi-scale metabolic modeling. Metabolic models of metabolism are mathematical, network-based, and large-scale representations of the set of chemical reactions for which various types of evidence exists. He is studying systems that can be modeled, model types, model applications, the development of new modeling techniques, and publicly available yet underleveraged datasets, along with potential collaborations within WSU and the PNNL. He aims to develop a broad program of research addressing key challenges such as improved plant tolerance to heat and drought stresses, drug repurposing, and designing cyanobacteria as CO2 to biochemical platforms. Schroeder hold a bachelor’s degree from Iowa State University and a PhD in Chemical and Biomolecular Engineering from the University of Nebraska, Lincoln. Before coming to WSU, he was a postdoctoral scholar at Pennsylvania State University.

Jesse Weaver

Instructor, Department of Civil and Environmental Engineering.

A WSU alumnus, Jesse Weaver has more than seven years of industry experience in the area of structural engineering and as a staff engineer at the Idaho Transportation Department. He will be teaching the department’s capstone, ethics, and geotechnical engineering courses and will serve as co-advisor for the WSU chapter of ASCE.

Huiyun Wu

Assistant Professor, Environmental Engineering, Department of Civil and Environmental Engineering.

Dr. Wu’s research is the area of microbial water equality, environmental microbiome, and water research. Her research is focused on data-driven strategies for advancing water sustainability, molecular microbiology applications, and environmental microbiome investigation. She has worked on multiple interdisciplinary research projects, including water reuse, wastewater-based epidemiology and sanitary sewage overflow survey, environmental metagenomics, microbial source tracking, stormwater management, and microbial water quality modeling. She is also interested in participating citizen science and serving the community. Before joining WSU, Dr. Wu participated in a two-year Oak Ridge Institute for Science and Education (ORISE) fellowship with the U.S. Environmental Protection Agency (EPA), and served as a Postdoctoral Fellow in the Department of Environmental Health Sciences in the School of Public Health & Tropical Medicine at Tulane University. She holds a Ph.D. in Environmental Engineering from Michigan State University.

Thesis Statement for Informative

Thesis statement generator for informative essay.

In the realm of informative essays, the thesis statement stands as a beacon of clarity, guiding readers to the heart of your insights. With the primary goal of educating the audience, the thesis must succinctly present the focal information you’re about to unpack. From setting the stage to presenting facts, every thesis matters. Dive deep into examples, unravel the crafting process, and learn valuable techniques to ensure your informative essay begins on the strongest note.

What is an Informative Essay Thesis Statement? – Definition

An informative essay thesis statement is a succinct declaration that outlines the primary focus or main point of the essay. Unlike argumentative essays, which make a claim and aim to persuade, an informative essay thesis seeks to educate the reader about a specific topic, providing a foundation for the detailed information that follows. The statement acts as a roadmap, giving the reader a glimpse of what to expect in the essay’s subsequent sections.

What is the Best Thesis Statement Example for Informative Essay?

While the “best” thesis statement often depends on the topic and the intent of the essay, here’s a generic example that embodies the characteristics of a strong informative thesis:

“The metamorphosis of a butterfly, from a humble caterpillar to a resplendent winged creature, is a captivating process involving four distinct stages: egg, larva, pupa, and adult.”

This thesis provides a clear, concise overview of the essay’s topic and what the reader can expect to learn from the ensuing paragraphs.

100 Thesis Statement Examples for Informative Essay

Crafting a robust thesis for an informative essay is essential to guide your readers through the nuances of the topic you’re exploring. Your thesis acts as a lens, focusing the reader’s attention on the key aspects you’ll delve into. Below is a collection of meticulously curated thesis statements for informative essays, designed to inspire and guide your writing process.

“The history of the printing press revolutionized human communication, transforming societal structures and information dissemination.”
“Solar energy, derived from the sun’s rays, offers a sustainable and environmentally friendly power source, with numerous applications in modern society.”
“The water cycle is a continuous process, consisting of evaporation, condensation, precipitation, and collection, crucial to Earth’s climate and ecosystem.”
“Ancient Egyptian mummification was a detailed ritual, embodying religious beliefs, preparations for the afterlife, and sophisticated preservation techniques.”
“Quantum mechanics delves into the behavior of subatomic particles, challenging traditional physics and introducing concepts like superposition and entanglement.”
“The Great Barrier Reef, the world’s largest coral reef system, boasts biodiversity, faces environmental threats, and is crucial for global marine ecology.”
“Yoga, originating from ancient India, promotes physical, mental, and spiritual wellness, with various forms tailored to different needs.”
“The Silk Road was a vast network of trade routes connecting Asia and Europe, facilitating the exchange of goods, cultures, and ideas.”
“Chocolates, beyond being a delightful treat, have a rich history, production process, and health benefits when consumed in moderation.”
“Mental health is a multifaceted topic, encompassing emotional, psychological, and social well-being, with various factors influencing one’s mental state.”
“Leonardo da Vinci, a Renaissance polymath, contributed to art, science, and engineering, with masterpieces like the Mona Lisa and inventions ahead of his time.”
“Pandas, native to China, play a significant role in global conservation efforts due to their endangered status and ecological importance.”
“Photography, since its inception, has evolved in techniques and styles, influencing society’s perception of reality and memory.”
“Green architecture integrates eco-friendly materials and energy-efficient designs to minimize environmental impact and promote sustainability.”
“Sleep is a vital physiological process, with stages like REM and deep sleep, affecting cognitive function, mood, and overall health.”
“Origami, the Japanese art of paper folding, has cultural significance, mathematical principles, and therapeutic benefits.”
“The evolution of human language encompasses physiological changes, societal developments, and the emergence of linguistic diversity.”
“The Internet, from ARPANET to today’s global network, has transformed communication, business, and entertainment, shaping the modern world.”
“Black holes, mysterious cosmic entities, are regions of spacetime exhibiting gravitational forces from which nothing can escape, not even light.”
“Jazz, an original American art form, draws from various music styles, influencing culture, civil rights movements, and global music scenes.”
“Vaccination, a cornerstone of modern medicine, employs weakened or inactivated germs to train the immune system against diseases.”
“Greek mythology, a rich tapestry of gods, heroes, and monsters, played a central role in ancient Greek religion and culture.”
“Artificial intelligence, the simulation of human intelligence in machines, has applications in healthcare, finance, and more, heralding a new technological age.”
“Mount Everest, the world’s highest peak, has a complex geology, history of expeditions, and challenges related to climbing and environmental conservation.”
“Ballet, a classical dance form, has evolved over centuries, boasting different styles, techniques, and a profound impact on global dance culture.”
“Mars, the fourth planet from the sun, is a focus of space exploration, with studies on its atmosphere, geology, and potential for life.”
“The Amazon Rainforest, Earth’s largest tropical rainforest, houses unparalleled biodiversity and plays a pivotal role in the global climate system.”
“The human brain, a marvel of evolution, is responsible for cognition, emotion, and consciousness, with regions dedicated to specific functions.”
“The French Revolution, a tumultuous period in history, brought about political, social, and economic upheavals, shaping modern democracy.”
“The Grand Canyon, carved by the Colorado River, showcases layers of Earth’s history, geology, and offers a haven for biodiversity.”
“Hydroponics, a method of growing plants without soil, utilizes nutrient-rich water, offering solutions for urban farming and food scarcity.”
“The Mona Lisa, beyond its fame as a painting, carries stories of its creation, theft, and cultural significance in art history.”
“Quantum computing harnesses principles of quantum mechanics, promising breakthroughs in processing speed, cryptography, and complex problem-solving.”
“The phenomenon of bioluminescence, seen in various marine creatures, is a chemical reaction that produces light, aiding in camouflage, prey attraction, and communication.”
“The pyramids of Egypt, marvels of ancient engineering, were built as tombs for pharaohs, reflecting the civilization’s religious beliefs and technological prowess.”
“Nanotechnology, the manipulation of matter on an atomic scale, holds promise for medical treatments, electronics, and materials science.”
“The Roaring Twenties, a decade post-WWI, were marked by cultural shifts, economic prosperity, jazz, and the onset of the Great Depression.”
“Sushi, a staple of Japanese cuisine, has a history spanning centuries, varying styles, and a globalized presence in today’s culinary landscape.”
“Vincent van Gogh, though tormented in life, produced masterpieces like ‘Starry Night’, influencing modern art with his unique style and technique.”
“The concept of zero, integral to mathematics, originated from ancient civilizations, influencing arithmetic, algebra, and our understanding of the universe.”
“Biodiversity, the variety of life on Earth, is vital for ecosystem stability, human survival, and indicates the planet’s health.”
“The Industrial Revolution marked a shift from agrarian societies to industrial urban centers, revolutionizing technology, society, and the global economy.”
“Volcanoes, nature’s fiery vents, have diverse types and formation processes, playing roles in Earth’s geology and influencing climates.”
“The human genome, our genetic blueprint, has been mapped, offering insights into genetics, evolution, and potentials for personalized medicine.”
“Shakespeare’s works, from tragedies to comedies, offer insights into human nature, love, power, and have profoundly influenced literature and language.”
“Acupuncture, an ancient Chinese therapy, involves inserting needles at specific points to balance energy and treat various ailments.”
“The Antarctic, a frozen frontier, is crucial for climate research, housing unique ecosystems and holding mysteries beneath its ice.”
“Meditation, a practice of focused attention, offers benefits like stress reduction, improved cognition, and greater self-awareness.”
“The Periodic Table organizes chemical elements based on atomic number, guiding our understanding of chemistry, reactions, and element properties.”
“The concept of time, from sundials to atomic clocks, has been central to human civilizations, influencing cultures, sciences, and daily life.
“Gut microbiome, the community of microorganisms living in our intestines, has profound implications on our health, mood, and disease susceptibility.”
“The Renaissance, spanning the 14th to 17th century, marked a cultural awakening in art, science, and thought, laying the foundation for the modern world.”
“Artificial neural networks, inspired by the human brain, form the basis of deep learning, propelling advancements in image recognition, language translation, and more.”
“The concept of relativity, introduced by Einstein, transformed our understanding of time, space, and the universe, challenging Newtonian physics.”
“The cultural and religious festival of Diwali, celebrated predominantly in India, signifies the triumph of light over darkness and good over evil.”
“J.R.R. Tolkien’s ‘The Lord of the Rings’ not only narrates an epic tale of heroism but delves deep into themes of friendship, power, and corruption.”
“Climate change, driven primarily by human activities, has far-reaching consequences on weather patterns, sea levels, and global ecosystems.”
“Impressionism, an art movement in the 19th century, captures fleeting moments with loose brushwork, championed by artists like Monet and Renoir.”
“Holography, the science of producing three-dimensional images, has applications in medicine, art, and data storage, promising future advancements.”
“The discovery of DNA’s double helix structure by Watson and Crick revolutionized biology, paving the way for genetic research and biotechnological innovations.”
“Coffee, beyond a popular beverage, has a rich history of cultivation, trade, and cultural significance across continents.”
“Migration patterns of monarch butterflies, traveling thousands of miles, are a remarkable phenomenon of nature, influenced by environmental cues and genetic factors.”
“The Roman Empire, with its vast territories and lasting legacies, has impacted modern governance, architecture, and language.”
“Virtual reality, an immersive technology, has transcended gaming to find applications in medicine, education, and real estate.”
“Dream analysis, rooted in psychological theories of Freud and Jung, delves into the subconscious mind, interpreting symbols and emotions for insights.”
“Beekeeping, an age-old practice, supports biodiversity, provides honey, and plays a crucial role in global food production through pollination.”
“The concept of black markets, operating outside sanctioned channels, impacts global economies, ethics, and law enforcement challenges.”
“The evolution of music, from classical symphonies to contemporary genres, reflects societal changes, technological innovations, and cultural exchanges.”
“Neuroplasticity, the brain’s ability to reorganize and adapt, challenges previous beliefs about brain rigidity and offers hope for injury recovery.”
“Taj Mahal, an architectural marvel in India, stands as a testament to eternal love, Mughal artistry, and intricate craftsmanship.
“The Silk Road, not just a trade route, fostered cultural exchanges, spread religions, and laid the groundwork for globalization in the ancient world.”
“Telecommunication, with its evolution from telegraphs to smartphones, has reshaped society, influencing communication habits, businesses, and global connectedness.”
“Veganism, beyond a dietary choice, carries implications for animal rights, environmental sustainability, and global food resources.”
“The architecture of Gaudi, particularly in Barcelona, embodies a unique blend of nature, religion, and modernism, attracting millions of admirers worldwide.”
“Galaxies, vast cosmic structures containing billions of stars, provide insights into the universe’s formation, dark matter, and the fate of cosmic bodies.”
“Procrastination, more than just delaying tasks, is a complex psychological behavior with implications for productivity, mental health, and personal growth.”
“Jazz, birthed in New Orleans, embodies improvisation and cultural synthesis, influencing numerous other genres and reflecting societal changes.”
“The Great Wall of China, beyond a monumental feat of engineering, symbolizes the lengths to which societies will go to defend their beliefs and territories.”
“Human rights, a universal framework for dignity and equality, have evolved over centuries, shaping global policies, revolutions, and societal values.”
“Pandemics, from the Black Plague to COVID-19, have shifted the course of history, influencing medical advancements, societal structures, and global economies.”
“Cryptocurrency, decentralized digital money, challenges traditional banking systems, offering potential for financial freedom but also sparking debates on regulation.”
“The Amazon Rainforest, often termed the ‘lungs of Earth’, plays a critical role in global climate regulation, biodiversity, and indigenous cultures.”
“The Eiffel Tower, initially criticized but now an icon of France, represents engineering prowess, national pride, and the changing tides of public opinion.”
“Ballet, a disciplined art form with roots in the Italian Renaissance, conveys stories, emotions, and has evolved with cultural and societal shifts.”
“The concept of infinity, both in mathematics and philosophy, challenges human comprehension and has led to profound discoveries and existential debates.”
“The Grand Canyon, carved by the Colorado River, stands as a testament to nature’s power and the geological history of Earth.”
“Storytelling, intrinsic to human culture, serves various purposes, from passing down traditions to marketing brands in the modern age.”
“Yoga, beyond physical postures, is an ancient practice promoting holistic well-being, spiritual growth, and mental clarity.”
“The Louvre Museum, housing thousands of artworks, narrates a history of art, culture, and the evolution of human civilization.”
“Photography, from daguerreotypes to digital, captures moments in time, influencing art, journalism, and how society perceives reality.”
“Mount Everest, standing as the highest peak, isn’t just a mountaineer’s challenge but a symbol of human perseverance and our relationship with nature.”
“Mars exploration, beyond the realm of science fiction, provides insights into planetary evolution, life beyond Earth, and the future of human space colonization.”
“Coral reefs, often called the rainforests of the sea, are vibrant ecosystems, vital to marine life, coastal economies, and indicate global climate health.”
“Shakespeare’s ‘Hamlet’ doesn’t merely tell a tale of revenge but delves deep into themes of existentialism, morality, and the human psyche.”
“Quantum mechanics, a foundation of modern physics, challenges classical notions, introducing concepts like superposition and entanglement, reshaping our understanding of reality.”
“The Pyramids of Giza, not just architectural marvels, offer insights into ancient Egyptian beliefs, astronomical knowledge, and societal organization.”
“Hydrogen as an energy source, while in its infancy, holds potential to revolutionize the energy sector, offering a cleaner alternative to fossil fuels.”
“The cultural phenomenon of Anime, originating in Japan, transcends entertainment, reflecting societal issues, personal identities, and diverse genres of storytelling.”
“Meditation, rooted in ancient traditions, serves as a tool for mental well-being, stress relief, and cognitive enhancement in our fast-paced modern world.”
“The French Revolution, while a bloody period, led to the overthrow of monarchy, shaping modern political ideologies, rights, and global democratic movements.

An informative essay thesis statement is a condensed form of your essay’s primary argument, serving as a roadmap for your readers. The process of developing such a statement requires synthesizing the main idea of your topic and presenting it in a concise manner to captivate and inform your audience from the beginning.

How do you write a thesis for an informative essay? – Step by Step Guide

Understand the Prompt : Before you can create a thesis, understand the prompt or the topic you’re addressing. This ensures your thesis aligns with what you are expected to write about.
Research Thoroughly : Dive deep into your topic. Gather all necessary details, facts, and data that will help you get a comprehensive view of the subject.
Identify the Main Idea : What is the primary message or insight you want your readers to grasp? This will form the core of your thesis.
Keep it Specific : Your thesis should not be overly broad. Instead, focus on a specific aspect of the topic that your essay will explore.
Make it Clear and Concise : Your thesis statement shouldn’t be a complex sentence. It should be clear, direct, and easy for the reader to understand.
Avoid Opinions : An informative essay provides information and insight. It doesn’t try to persuade the reader or present the writer’s personal opinion.
Review and Refine : After drafting your thesis, read it aloud. Does it flow? Is it clear? Make necessary revisions until it fits your essay’s scope and direction perfectly.

Tips for Writing an Informative Essay Thesis Statement

Stay Neutral : Your thesis shouldn’t convey bias or opinion. Stick to facts and neutral language.
Position it Right : Traditionally, the thesis statement is positioned at the end of the introduction to guide the reader into the main body.
Stay Focused : Your thesis should be specific to the points you’ll be making in your essay. If a point doesn’t support your thesis, consider removing it from your essay.
Seek Feedback : Before finalizing your thesis, seek feedback. Fresh eyes can offer valuable insights and catch inconsistencies.
Revisit After Writing : Once your essay is complete, revisit your thesis. Does your essay deliver what your thesis promises? If not, tweak it so that it aligns with your essay’s content.

Crafting a compelling thesis for an informative essay is a balancing act between providing clear, concise information and sparking curiosity in readers. By following the aforementioned steps and tips, writers can guide their audience seamlessly through the information while ensuring comprehension and interest.

Text prompt

Instructive
Professional

Write a Thesis Statement for Informative Essay on the process of recycling plastics.

Create a Thesis Statement for Informative Essay about the history of the internet.

IMAGES

(PDF) Presentation of: The role of artificial intelligence in
(PDF) Artificial Intelligence Aided Architectural Design
(PDF) Application of Artificial Intelligence in Architectural Design
(PDF) Architecture, urban design and Artificial Intelligence (AI
(PDF) AIA Artificial Intelligence in Architecture GENERAL UNDERSTANDING
Pin on Architectural Thesis Award

VIDEO

AI Rendering for Architects
Architectural concepts study using AI
How do I write my PhD thesis about Artificial Intelligence, Machine Learning and Robust Clustering?
ScholarWriterAI
Forest house architecture Futuristic house drawn by AI
CSS Essay Writing Techniques || CSS Essay on Artificial Intelligence || MaoozRiaz

COMMENTS

The impact of introducing artificial intelligence technology to
This thesis, essentially, is an exploration of the ways that "Artificial Intelligence" techniques may support systematic and rational architectural design and, by extension, the "Building Systems" process. The motivation for working in this area of research stems from the serious need to develop a new methodological design approach for architects.
The Impact of Artificial intelligence on the future of architecture
Innovations in computing technologies became a large part of fulfilling everyday needs. Artificial intelligence machines are advancing beyond machine-like abilities, which enables them to process ...
Architecture+AI -The impact of AI in Architecture
Ajit Clement. "This dissertation explores the impact of artificial intelligence (AI) on the field of architecture. Through a review of literature and case studies, the study examines the ways in which AI is being used in building information modeling (BIM), building performance optimization, data analysis, and generative design.
PDF Artificial Intelligence in Architecture and its Impact on Design ...
Findings - The thesis contributes to understanding how new technology such as AI can affect creativity in the design process. It explored how the creative process is currently structured and how it will be affected by the implementation of AI. It provides an overview ... The . artificial intelligence, - architecture) ...
Artificial intelligence in architecture
Artificial intelligence, and in particular machine learning, is a fast-emerging field. Research on artificial intelligence focuses mainly on image-, text- and voice-based applications, leading to breakthrough developments in self-driving cars, voice recognition algorithms and recommendation systems. In this article, we present the research of ...
Generative AI for Architectural Design: A Literature Review
Generative Artificial Intelligence (AI) has pioneered new methodological paradigms in architectural design, significantly expanding the innovative potential and efficiency of the design process. This paper explores the extensive applications of generative AI technologies in architectural design, a trend that has benefited from the rapid development of deep generative models. This article ...
Artificial intelligence applied to conceptual design. A review of its
In this work, we will examine the main research projects that applied artificial intelligence solutions to the design of form in architecture. As early as 1987, Soddu [ 7 ] created artificial DNA of Italian medieval cities which he used to define the Generative Design approach to Architecture and City Design in his book "Citta' Aleatorie."
ArchiGAN: Artificial Intelligence x Architecture
Master thesis: "Enabling alternative architectures: Collaborative frameworks for participatory design." Cambridge, MA: Harvard Graduate School of Design. Google Scholar Martinez, N. (2016). Suggestive drawing among human and artificial intelligences. Cambridge, MA: Harvard Graduate School of Design.
ArchiGAN: Artificial Intelligence x Architecture
As described in Fig. 3, each model of the stack handles a specific task of the workflow: (I) footprint massing, (II) program repartition, (III) furniture layout. An architect is able to modify or fine-tune the model's output between each step, thereby achieving the expected machine-human interaction. Fig. 3 Generation stack in three models.
AI + Architecture
Artificial Intelligence Architecture Data Science Generative Adversarial Networks. Artificial Intelligence, as a discipline, has already been permeating countless fields, bringing means and methods to previously unresolved challenges, across industries. The advent of AI in Architecture is still in its early days but offers.
Architecture in the Age of Artificial Intelligence: An introduction to
The first volume in a two-book series, Architecture in the Age of Artificial Intelligence introduces AI for designers and considers its positive potential for the future of architecture and design ...
Artificial intelligence in architecture: Generating conceptual design
Using DNNs in spatial design is more complex. Architecture is shaped by a broad set of interdependent issues. In his treatise De architectura, written in 80 BC, Vitruvius wrote that any successful architecture should provide for function, beauty and structure.And, Walter Gropius in Scope of Total Architecture claimed that 'good architecture should be a projection of life itself that implies ...
Use of Artificial Intelligence in the Field of Sustainable Architecture
Research objective. One of the current topics of application of information technolo-. gy (IT) to support architectural d esign and optimization of build-. ing operation is the use of artificial ...
PDF Impact of Artificial Intelligence in Design
1.4 Thesis structure The structure of the thesis comprises five primary sections: Introduction, Overview of Artificial Intelligence, Impact of AI in Design, Research Methodology, Results, and Discussion & Conclusion. The first section explains the thesis background information, outlines the objectives and the research questions.
The role of Artificial Intelligence in architectural design
Leaving aside the ethical considerations related to artificial intelligence in general, it is necessary to reflect on the data we inject in these processes. The ways we collect and organize them can lead to biased output, and on more than one occasion artificial intelligence has shown discriminatory behaviours concerning ethnicity or gender [8].
Pioneers: 6 Practices Bringing AI into Architecture
Artificial Intelligence in Architecture. Image Courtesy of 3XN. Although this is solely a research project, the example shows how a renowned architecture practice positions itself with regards to ...
PDF Artificial Intelligence and Machine Learning: Current Applications in
This thesis will define machine learning and artificial intelligence for the investor and real estate audience, examine the ways in which these new analytic, predictive, and automating technologies are being used in the real estate industry, and postulate potential
FIU Libraries: Artificial Intelligence: Dissertations & Theses
Many universities provide full-text access to their dissertations via a digital repository. If you know the title of a particular dissertation or thesis, try doing a Google search. OATD (Open Access Theses and Dissertations) Aims to be the best possible resource for finding open access graduate theses and dissertations published around the world with metadata from over 800 colleges ...
Enhancing Architectural Design with Artificial Intelligence: A ...
I n the ever-evolving field of architectural design, artificial intelligence (AI) has emerged as a transformative force that promises to reshape the way architects approach their work. While some ...
The Transformative Power of Artificial Intelligence: A Deep Dive into
Understanding Artificial Intelligence. Artificial Intelligence refers to the simulation of human intelligence in machines that are programmed to think and learn. These systems can perform tasks that typically require human intelligence, such as recognizing speech, making decisions, and translating languages.
AIA Artificial Intelligence in Architecture GENERAL UNDERSTANDING AND
Deep Learning is used by Google in its voice and image recognition algorithms, by Netflix and Amazon to decide what you want to watch or buy next, and by researchers at MIT to predict the future. (Marr, 2016) 3 HOW IS AI HELPING IN ARCHITECTURE 2. HOW IS AI HELPING IN ARCHITECTURE Artificial Intelligence is using data to change the world.
Artificial Intelligence
Final Exam (Thesis Defense) Ph.D. / M.S. Thesis Format Review Guidelines; MS Program; ... Architecture, Compilers, and Parallel Computing; Artificial Intelligence; ... artificial intelligence includes several key areas where our faculty are recognized leaders: computer vision, machine listening, natural language processing, machine learning and ...
Doctoral Thesis: Parameterizations of Neural Fields
Artificial Intelligence and Decision-making combines intellectual traditions from across computer science and electrical engineering to develop techniques for the analysis and synthesis of systems that interact with an external world via perception, communication, and action; while also learning, making decisions and adapting to a changing environment.
Frontiers
As a result, Explainable Artificial Intelligence (XAI) has emerged as a relevant research field aiming to develop methods and techniques that allow stakeholders to understand the inner workings and outcome of deep learning-based systems (Gunning et al., 2019; Arrieta et al., 2020).
The Top Artificial Intelligence Colleges in Texas
Texas A&M University offers a campus-based, hybrid or online Master of Science in Artificial Intelligence. The hybrid option allows students to take some classes online, and some on-campus. Credits: 34 - 37. There are non-thesis and thesis options available. Students can choose from four emphases: Computational Linguistics; Psychology; Computer ...
(PDF) Application of Artificial Intelligence in Architectural Design
The world of architecture has not remained external to this phenomenon: many researchers are working on the applications of artificial intelligence to architectural design, a few design software allow exploiting machine learning algorithms, and some large architectural firms have begun to experiment with deep learning methods to put into ...
Voiland College Welcomes New Faculty for 2024-25
She holds a Master of Architecture and a Master of Urban and Community design from the University of South Florida's School of Architecture and Community Design. ... trustworthy artificial intelligence, and IoT. ... (EPFL) in 2018 and 2023, respectively. He received the EPFL PhD Thesis Distinction in Electrical Engineering in 2023 for his ...
Thesis Statement for Informative
"Artificial intelligence, the simulation of human intelligence in machines, has applications in healthcare, finance, and more, heralding a new technological age." "Mount Everest, the world's highest peak, has a complex geology, history of expeditions, and challenges related to climbing and environmental conservation."

Computer Science > Machine Learning

Submission history

References & Citations

BibTeX formatted citation

Bibliographic and Citation Tools

arXivLabs: experimental projects with community collaborators

ArchiGAN: Artificial Intelligence x Architecture

Cite this chapter

Access this chapter

Similar content being viewed by others

Big Architect, Little Architect

AI art in architecture

Author information

Corresponding author

Editor information

Rights and permissions

Copyright information

About this chapter

Download citation

Share this chapter

Use of Artificial Intelligence in the Field of Sustainable Architecture: Current Knowledge

Abstract and Figures

Pioneers: 6 Practices Bringing AI into Architecture

AI+ Architecture / Stanislas Chaillou

Artificial Intelligence in Architecture / 3XN

Daedalus Pavilion / Ai Build

Plaza Life Revisited / XL Lab SWA Group

Generative Design Tool / Sidewalk Labs

Ada / Jenny Sabin

Image gallery

想阅读文章的中文版本吗?

科技建筑先驱者：6家引入人工智能的事务所

Artificial Intelligence

FIU dissertations

Non-FIU dissertations

Related Sites

Information

Hubert Library, Biscayne Bay Campus

Directions: Green Library, MMC

Directions: Hubert Library, BBC

Please enter your credentials

The Transformative Power of Artificial Intelligence: A Deep Dive into AI

Understanding Artificial Intelligence

Core Components of AI

Applications of AI

1. Healthcare

3. Transportation

4. Entertainment

Ethical Considerations in AI

The Future of AI

The Importance of Employee Well-Being in Today’s Workplace

The Role and Importance of Information Technology (I.T.) in Today’s World

The Rise of ChatGPT and AI: Transforming Communication and Beyond

The Transformative Impact of Artificial Intelligence: Applications and Benefits

The Impact of AI on Modern Society

ChatGPT: Revolutionizing Human-Computer Interaction

Contact Us Atlantic International University

Related News

Strengths and Impact

Research Efforts and Groups

Faculty & Affiliate Faculty

Adjunct Faculty

Doctoral Thesis: Parameterizations of Neural Fields

Additional Location Details:

REVIEW article

1 Introduction

2 Rationalization methods for text classification

2.1 Extractive rationalization

2.1.1 Extractive methods

2.1.1.1 Multi-task models

2.1.1.2 Pipelined models

2.1.2 Attention-based methods

2.1.2.1 Soft-scores models

2.2 Abstractive rationale generation

2.2.1 Text-to-text methods

2.2.1.1 Multi-task models

2.2.2 Generative methods

2.2.2.1 Pipelined models

2.2.2.2 Multi-task models

3 Rationale-annotated datasets