Rotate your Networks: Better Weight Consolidation and Less Catastrophic Forgetting

TL;DR

This paper introduces a network reparameterization technique that improves elastic weight consolidation, significantly reducing catastrophic forgetting in lifelong learning across multiple datasets.

Contribution

It proposes a novel reparameterization method that approximately diagonalizes the Fisher Information Matrix, enhancing EWC's effectiveness in sequential task learning.

Findings

Significantly improved performance over standard EWC.

Achieved competitive results with state-of-the-art lifelong learning methods.

Effective across diverse datasets like MNIST, CIFAR-100, CUB-200, and Stanford-40.

Abstract

In this paper we propose an approach to avoiding catastrophic forgetting in sequential task learning scenarios. Our technique is based on a network reparameterization that approximately diagonalizes the Fisher Information Matrix of the network parameters. This reparameterization takes the form of a factorized rotation of parameter space which, when used in conjunction with Elastic Weight Consolidation (which assumes a diagonal Fisher Information Matrix), leads to significantly better performance on lifelong learning of sequential tasks. Experimental results on the MNIST, CIFAR-100, CUB-200 and Stanford-40 datasets demonstrate that we significantly improve the results of standard elastic weight consolidation, and that we obtain competitive results when compared to other state-of-the-art in lifelong learning without forgetting.