Modular-Relatedness for Continual Learning
This work addresses the problem of catastrophic forgetting for sequential task learners in continual learning, offering an incremental improvement by enhancing existing methods like EWC and GEM.
The paper tackles catastrophic forgetting in continual learning by proposing a technique that automatically extracts modular parts of neural networks and estimates task relatedness, resulting in improved retained accuracy and robustness to forgetting, especially with limited memory budgets.
In this paper, we propose a continual learning (CL) technique that is beneficial to sequential task learners by improving their retained accuracy and reducing catastrophic forgetting. The principal target of our approach is the automatic extraction of modular parts of the neural network and then estimating the relatedness between the tasks given these modular components. This technique is applicable to different families of CL methods such as regularization-based (e.g., the Elastic Weight Consolidation) or the rehearsal-based (e.g., the Gradient Episodic Memory) approaches where episodic memory is needed. Empirical results demonstrate remarkable performance gain (in terms of robustness to forgetting) for methods such as EWC and GEM based on our technique, especially when the memory budget is very limited.