Article Source

• Continual Learning Beyond Catastrophic Forgetting in Class-Incremental Scenarios

Continual Learning Beyond Catastrophic Forgetting in Class-Incremental Scenarios

• Speakers: Antonio Carta and Vincenzo Lomonaco

Abstract

Continual learning methods are evaluated on various objectives, such as reducing forgetting, improving learning new tasks, and computational efficiency. However, most of the literature on continual learning focuses on mitigating catastrophic forgetting in class-incremental learning (CIL), which is a limited and unrealistic problem. Fortunately, there are alternative evaluation methods and benchmarks that are simpler to use and closer to the spirit of lifelong learning. This tutorial provides an overview of continual learning, focusing on recent proposals for benchmarks and evaluation methodologies and concluding with open research questions.

Part I

Bio

Antonio Carta is an assistant professor at the University of Pisa and a member of the Pervasive AI Lab. His research focuses on continual learning in rehearsal-free and distributed/multi-agent settings. He is also the lead maintainer of Avalanche, an open-source continual learning library developed by ContinualAI. He has been an organizer of CLVISION22 and PAIW22, and a program committee member of ICML, CoLLAs, IJCNN/WCCI. Vincenzo Lomonaco is an Assistant Professor at the University of Pisa, Italy where he teaches the Artificial Intelligence master degree course. Currently, he also serves as Co-Founding President at ContinualAI, a non-profit research organization and the largest open community on Continual Learning for AI. Vincenzo is a Task Leader of two main European projects and a Principal Investigator of several industrial research contracts with companies such as Meta, Intel, Leonardo s.p.a. and SeaVision s.r.l.

Part II