Riemannian Walk for Incremental Learning: Understanding Forgetting and Intransigence

TL;DR

This paper introduces RWalk, a new incremental learning algorithm that addresses forgetting and intransigence, providing better accuracy and a balanced trade-off between these issues through novel metrics and theoretical insights.

Contribution

It proposes RWalk, a generalized approach building on EWC++ and Path Integral, with new metrics for forgetting and intransigence, and offers a comprehensive analysis of IL algorithms.

Findings

RWalk outperforms existing methods in accuracy on MNIST and CIFAR-100.

The new metrics effectively quantify forgetting and intransigence.

RWalk achieves a better balance between knowledge retention and updating.

Abstract

Incremental learning (IL) has received a lot of attention recently, however, the literature lacks a precise problem definition, proper evaluation settings, and metrics tailored specifically for the IL problem. One of the main objectives of this work is to fill these gaps so as to provide a common ground for better understanding of IL. The main challenge for an IL algorithm is to update the classifier whilst preserving existing knowledge. We observe that, in addition to forgetting, a known issue while preserving knowledge, IL also suffers from a problem we call intransigence, inability of a model to update its knowledge. We introduce two metrics to quantify forgetting and intransigence that allow us to understand, analyse, and gain better insights into the behaviour of IL algorithms. We present RWalk, a generalization of EWC++ (our efficient version of EWC [Kirkpatrick2016EWC]) and Path Integral [Zenke2017Continual] with a theoretically grounded KL-divergence based perspective. We provide a thorough analysis of various IL algorithms on MNIST and CIFAR-100 datasets. In these experiments, RWalk obtains superior results in terms of accuracy, and also provides a better trade-off between forgetting and intransigence.