Self-distilled Knowledge Delegator for Exemplar-free Class Incremental Learning

TL;DR

This paper introduces a knowledge delegator that uses self-distillation to transfer knowledge from a trained model to a new one, effectively mitigating catastrophic forgetting in exemplar-free class incremental learning.

Contribution

It proposes a novel knowledge delegator mechanism that enables exemplar-free incremental learning by transferring knowledge via self-distillation, outperforming existing methods.

Findings

Outperforms existing exemplar-free methods on four benchmarks.

Achieves performance comparable to exemplar-based methods without using exemplars.

Effective in maintaining old task performance in incremental learning.

Abstract

Exemplar-free incremental learning is extremely challenging due to inaccessibility of data from old tasks. In this paper, we attempt to exploit the knowledge encoded in a previously trained classification model to handle the catastrophic forgetting problem in continual learning. Specifically, we introduce a so-called knowledge delegator, which is capable of transferring knowledge from the trained model to a randomly re-initialized new model by generating informative samples. Given the previous model only, the delegator is effectively learned using a self-distillation mechanism in a data-free manner. The knowledge extracted by the delegator is then utilized to maintain the performance of the model on old tasks in incremental learning. This simple incremental learning framework surpasses existing exemplar-free methods by a large margin on four widely used class incremental benchmarks, namely CIFAR-100, ImageNet-Subset, Caltech-101 and Flowers-102. Notably, we achieve comparable performance to some exemplar-based methods without accessing any exemplars.