Home Instructors Contact Us


Tutorial: Addressing Bias and Hallucination in Large Language Models

Time and Venue

Abstract

In the landscape of natural language processing (NLP), addressing the challenges of bias and hallucination is paramount to ensuring the ethical and unbiased development of Large Language Models (LLMs). This tutorial delves into the intricate dimensions of LLMs, shedding light on the critical importance of understanding and mitigating the profound impacts of bias and hallucination. The tutorial begins with discussions on the complexity of bias propagation in LLM development, where we dissect its origins and far-reaching impacts along with the automatic evaluation metrics for bias measurement. We then present innovative methodologies for mitigating diverse forms of bias, including both static and contextualized word embeddings and robust benchmarking strategies. In addition, the tutorial explores the interlinkage between hallucination and bias in LLMs by shedding light on how bias can be perceived as a hallucination problem. Furthermore, we also talk about cognitively-inspired deep learning frameworks for hallucination detection which leverages human gaze behavior. Ultimately, this cutting-edge tutorial serves as a guiding light, equipping participants with indispensable tools and insights to navigate the ethical complexities of LLMs, thus paving the way for the development of unbiased and ethically robust NLP systems.

Tutorial Prerequisite

We intend to make the tutorial self-contained. The tutorial materials such as the slides and video recordings will published for later reference. Further reading materials beyond the content of this tutorial will be provided in the slides itself.

Instructors' Bio

  • Nihar Ranjan Sahoo is a PhD student in the Computer Science department of IIT Bombay, supervised by Prof. Pushpak Bhattacharyya. His research interest lies in Ethical AI, social biases/toxicity in languages, and explainability in NLP. He has given a tutorial on social bias detection and mitigation in NLP at ICON. He has published papers on bias detection at conferences such as BMVC, LREC, CoNLL, NAACL, AAAI, ACL.
  • Ashita Saxena is a 3rd year MS by Research (CSE) student at IIT Bombay guided by Prof. Pushpak Bhattacharyya. Her research focuses on hallucination detection and mitigation in NLP tasks and her work is published in EMNLP. She has worked as a Research Intern at IBM Research on Natural Language Generation (NLG).
  • Kishan Maharaj s an MS (by Research) student at IIT Bombay (CSE), guided by Prof. Pushpak Bhattacharyya. His research focuses on cognitively inspired natural language processing, specifically hallucination detection and mitigation. His work was published in EMNLP. He is currently working with IBM research on prompt-based hallucination mitigation. Formerly, he worked with Turtle Mint and TATA Sons on various data science problems.
  • Arif Ahmad is currently in the final year of a BTech/MTech dual degree in Electrical Engineering and AI at IIT Bombay. He is working in the area of Fairness and Bias in NLP systems and Models, under the supervision of Prof. Pushpak Bhattacharyya at the CFILT Lab in IIT Bombay.
  • Dr. Abhijit Mishra an Assistant Professor of Practice at the School of Information, University of Texas at Austin, boasts extensive experience in ML and NLP, spanning over a decade. Formerly a Research Scientist at Apple Inc. and IBM Research, his contributions to NLP-based products like Siri and Watson are noteworthy. With notable publications at key AI and NLP conferences such as ACL, EMNLP, and AAAI, he has demonstrated expertise in various NLP domains, including multilingual and multimodal Natural Language Understanding and Generation, Sentiment Analysis, and Cognitive NLP with eye-tracking. Dr. Mishra’s recent focus on ethical LLM development aligns closely with the theme of the tutorial.
  • Dr. Pushpak Bhattacharyya s a Professor of Computer Science and Engineering at IIT Bombay. Educated in the IIT System (B.Tech IIT Kharagpur, M.Tech IIT Kanpur, PhD IIT Bombay), Dr. Bhattacharyya has done extensive research in Natural Language Processing and Machine Learning. He has published more than 350 research papers, has authored/co-authored 6 books including a textbook on machine translation, and has guided more than 350 students for their PhD, Masters and Under- graduate thesis. He has received many Research Excellence Awards- Manthan award from Ministry of IT, H.H. Mathur and P.K.Patwardhan awards from IIT Bombay, VNMM award from IIT Roorkee, and substantial research grants from Government and industry. Prof. Bhattacharyya holds the Bhagat Singh Rekhi Chair Professorship of IIT Bombay, is a Fellow of National Academy of Engineering, Abdul Kalam National Fellow, Distinguished Alumnus of IIT Kharagpur, past Director of IIT Patna and pastPresident of ACL.

Tutorial Slides

Slides