@channel This Friday [ 8-1-2021] 5.30pm CET , for the ContinualAI Reading Group , Mehdi Abbana Bennani (Aqemia) will present the paper:
Abstract: In Continual Learning settings, deep neural networks are prone to Catastrophic Forgetting. Orthogonal Gradient Descent was proposed to tackle the challenge. However, no theoretical guarantees have been proven yet. We present a theoretical framework to study Continual Learning algorithms in the Neural Tangent Kernel regime. This framework comprises closed form expression of the model through tasks and proxies for Transfer Learning, generalisation and tasks similarity. In this framework, we prove that OGD is robust to Catastrophic Forgetting then derive the first generalisation bound for SGD and OGD for Continual Learning. Finally, we study the limits of this framework in practice for OGD and highlight the importance of the Neural Tangent Kernel variation for Continual Learning with OGD.
The event will be moderated by: Vincenzo Lomonaco
Eventbrite Event (to save it on your calendar): Generalisation Guarantees for CL with Orthogonal Gradient Descent Tickets, Fri, Jan 8, 2021 at 5:30 PM | Eventbrite
Slides: ContinualAI Reading Group - Generalisation Guarantees for Continual Learning with Orthogonal Gradient Descent.pdf - Google Drive
YouTub recording: https://youtu.be/UOwgkBWJB54