Complementary Learning for Overcoming Catastrophic Forgetting Using Experience Replay
Mohammad Rostami, Soheil Kolouri, Praveen K. Pilly
TL;DR
This work tackles catastrophic forgetting in sequential multitask learning by embedding all tasks into a shared discriminative space and using a generative autoencoder to produce pseudo-data for experience replay. By modeling a common embedding distribution with a GMM and aligning current task representations to this distribution via a sliced Wasserstein discrepancy, the method mitigates forgetting without storing past data. Theoretical bounds grounded in optimal transport support the approach, and empirical results on permuted MNIST and related-domain digit tasks demonstrate reduced forgetting and effective knowledge integration across tasks. The CLEER framework offers a memory-efficient alternative to full replay and complements weight-consolidation methods for robust lifelong learning.
Abstract
Despite huge success, deep networks are unable to learn effectively in sequential multitask learning settings as they forget the past learned tasks after learning new tasks. Inspired from complementary learning systems theory, we address this challenge by learning a generative model that couples the current task to the past learned tasks through a discriminative embedding space. We learn an abstract level generative distribution in the embedding that allows the generation of data points to represent the experience. We sample from this distribution and utilize experience replay to avoid forgetting and simultaneously accumulate new knowledge to the abstract distribution in order to couple the current task with past experience. We demonstrate theoretically and empirically that our framework learns a distribution in the embedding that is shared across all task and as a result tackles catastrophic forgetting.