EXACFS -- A CIL Method to mitigate Catastrophic Forgetting

TL;DR

This paper proposes EXACFS, a novel class incremental learning method that uses feature significance estimation and distillation to reduce catastrophic forgetting in neural networks, showing improved stability and plasticity.

Contribution

Introduction of EXACFS, a new CIL approach that estimates feature significance via loss gradients and employs distillation to mitigate forgetting.

Findings

Outperforms existing methods on CIFAR-100 and ImageNet-100

Effectively balances stability and plasticity in continual learning

Reduces catastrophic forgetting significantly

Abstract

Deep neural networks (DNNS) excel at learning from static datasets but struggle with continual learning, where data arrives sequentially. Catastrophic forgetting, the phenomenon of forgetting previously learned knowledge, is a primary challenge. This paper introduces EXponentially Averaged Class-wise Feature Significance (EXACFS) to mitigate this issue in the class incremental learning (CIL) setting. By estimating the significance of model features for each learned class using loss gradients, gradually aging the significance through the incremental tasks and preserving the significant features through a distillation loss, EXACFS effectively balances remembering old knowledge (stability) and learning new knowledge (plasticity). Extensive experiments on CIFAR-100 and ImageNet-100 demonstrate EXACFS's superior performance in preserving stability while acquiring plasticity.