Continual Learning and Knowledge Sharing

Abstract: Real-world data is non-stationary and characterized by continuous changes. As a result, deep neural networks (DNN) must be adapted over time. Continual learning methods are designed to adapt DNNs by learning new tasks incrementally without forgetting old ones. In this talk, we will show how continual learning naturally arises from many applications and highlight its main challenges. Then, we will discuss recent research on knowledge sharing and explain how continual learning models can adapt more efficiently by sharing their task-specific knowledge.

Speaker Bio: Antonio Carta is an assistant professor at the University of Pisa and a member of the Pervasive AI Lab. His research is focused on continual learning and its applications to computer vision and time series. He is also the lead maintainer of Avalanche, a continual learning library based on pytorch, and a board member of ContinualAI, a no-profit organization focused on the dissemination of continual learning research.