Two Complementary Perspectives to Continual Learning: Ask Not Only What to Optimize, But Also How
This work addresses the problem of catastrophic forgetting in continual learning for AI systems, but it is incremental as it builds on existing methods without achieving clear empirical gains.
The paper identifies a 'stability gap' in continual learning where even perfect approximations of joint loss lead to temporary forgetting when switching tasks, and it argues for focusing on both optimization objectives and trajectories, though initial experiments combining these approaches did not yield consistent benefits.
Recent years have seen considerable progress in the continual training of deep neural networks, predominantly thanks to approaches that add replay or regularization terms to the loss function to approximate the joint loss over all tasks so far. However, we show that even with a perfect approximation to the joint loss, these approaches still suffer from temporary but substantial forgetting when starting to train on a new task. Motivated by this 'stability gap', we propose that continual learning strategies should focus not only on the optimization objective, but also on the way this objective is optimized. While there is some continual learning work that alters the optimization trajectory (e.g., using gradient projection techniques), this line of research is positioned as alternative to improving the optimization objective, while we argue it should be complementary. In search of empirical support for our proposition, we perform a series of pre-registered experiments combining replay-approximated joint objectives with gradient projection-based optimization routines. However, this first experimental attempt fails to show clear and consistent benefits. Nevertheless, our conceptual arguments, as well as some of our empirical results, demonstrate the distinctive importance of the optimization trajectory in continual learning, thereby opening up a new direction for continual learning research.