Towards guarantees for parameter isolation in continual learning
This work addresses the lack of theoretical guarantees in continual learning, which is a problem for researchers and practitioners seeking reliable sequential training of deep neural networks.
The paper tackles catastrophic forgetting in continual learning by analyzing the loss landscape geometry of neural networks, establishing provable guarantees against forgetting for parameter isolation methods.
Deep learning has proved to be a successful paradigm for solving many challenges in machine learning. However, deep neural networks fail when trained sequentially on multiple tasks, a shortcoming known as catastrophic forgetting in the continual learning literature. Despite a recent flourish of learning algorithms successfully addressing this problem, we find that provable guarantees against catastrophic forgetting are lacking. In this work, we study the relationship between learning and forgetting by looking at the geometry of neural networks' loss landscape. We offer a unifying perspective on a family of continual learning algorithms, namely methods based on parameter isolation, and we establish guarantees on catastrophic forgetting for some of them.