Gradient Correlation Subspace Learning against Catastrophic Forgetting

Efficient continual learning techniques have been a topic of significant research over the last few years. A fundamental problem with such learning is severe degradation of performance on previously learned tasks, known also as catastrophic forgetting. This paper introduces a novel method to reduce catastrophic forgetting in the context of incremental class learning called Gradient Correlation Subspace Learning (GCSL). The method detects a subspace of the weights that is least affected by previous tasks and projects the weights to train for the new task into said subspace. The method can be applied to one or more layers of a given network architectures and the size of the subspace used can be altered from layer to layer and task to task. Code will be available at https://github.com/vgthengane/GCSL Traditionally, as a neural network learns multi-class classification, the network learns to extract features indicative of the target labels and to perform the classification. When learning on all target labels simultaneously, networks are able to reach the highest accuracy presumably because they can learn the features relevant to all labels at the same time. If a network is trained on a set of labels and then at a later point trained on a different set of labels, the network often "catastrophically forgets" how to classify the original labels.