Скачать или смотреть Introduction to RLHF | PyImageSearch | Learn how ChatGPT works!

Souradip is currently a 2nd-year Ph.D. Computer Science Ph.D. student at the University of Maryland, College Park, working in the Foundations of Reinforcement Learning in Sequential Decision Making. He aims to develop large-scale robust algorithms for sequential decision-making tasks under practical and challenging limitations to make Safe, Fair, Robust, and Aligned to Human behavior & Preferences - bridge the Gap b/w Theory and Practice. Recently received the Outstanding Paper Award, TSRML at Neurips2022 and Outstanding Reviewer Awards, Neurips 2022, AISTATS 2023. As a part of the Ph.D. program, he has published in venues including ICML, Neurips, AAAI, CoRL, and ICRA. In the past, Souradip has worked for 3 years as a Research AI Scientist at Walmart Labs, India after completing my Masters from the Indian Statistical Institute in 2018 summa cum laude and also a Google Developers Expert in Machine Learning (2019). Co-authored several US patents and top-tier publications in the field of AI & ML applications in the NLP and Computer Vision domain as a part of Walmart Labs and GDE-ML.

The major success behind the exceptional performance of ChatGPT can be attributed to the Reinforcement Learning from Human Feedback which has significantly improved the performance of Language models. Aligning with Human Feedback is extremely critical in the current times in the context of Safety, Security, and Trustworthy AI. RLHF provides an efficient framework for alignment with only human preferences. In this session, Souradip will give an introduction to the RLHF framework and challenges and what are the next steps.