Séminaire ayant lieu dans le cadre du doctorat en informatique cognitive, en collaboration avec le centre de recherche CRIA et l'ISC    

Bing LIU

Jeudi le 27 février 2025 à 10h30

Local: PK-5115   (Il est possible d'y assister en virtuel en vous inscrivant ici)    

TITRE :    Achieving Upper Bound Accuracy in Continual Learning

RÉSUMÉ

The ability to continuously learn and accumulate knowledge over a lifetime is a hallmark of human intelligence. However, this essential capability is missing in current machine learning paradigms. This talk explores continual learning in machine learning, with a focus on the challenges of catastrophic forgetting and inter-task class separation. These issues have prevented existing methods from reaching the theoretical upper-bound performance, often with a significant gap. Our recent work demonstrates that achieving this upper bound is indeed possible, offering intriguing insights into both cognition and the foundations of AI.

BIOGRAPHIE

Bing LIU is a Distinguished Professor and Peter L. and Deborah K. Wexler Professor of Computing at the University of Illinois Chicago. He earned his Ph.D. from the University of Edinburgh. His current research interests include continual or lifelong learning, continual learning dialogue systems, sentiment analysis, machine learning, and natural language processing. He is the author of several books on these topics and has also received multiple Test-of-Time awards for his research papers. He is a Fellow of ACM, AAAI, and IEEE.

RÉFÉRENCES

Chen, Z., & Liu, B. (2018). Lifelong machine learning. Morgan & Claypool Publishers.

Ke, Z., Shao, Y., Lin, H., Konishi, T., Kim, G., & Bing Liu. Continual Pre-training of Language Models.  ICLR-2023.

Kim. G., Xiao, C., Konishi, T., Ke, Z., & Liu, B. A Theoretical Study on Solving Continual Learning. NeurIPS-2022.

Liu, B. (2023). Grounding for Artificial Intelligence. arXiv preprint arXiv:2312.09532.

Momeni, S., Mazumder, S., & Liu, B. Continual Learning Using a Kernel-Based Method Over Foundation Models, AAAI-2025, 2025.