FETCH: A Memory-Efficient Replay Approach for Continual Learning in Image Classification

TL;DR

FETCH introduces a two-stage compression method for replay in class-incremental continual learning, significantly reducing memory usage while maintaining high accuracy on CIFAR datasets, outperforming autoencoders.

Contribution

The paper proposes FETCH, a novel two-stage compression approach that enhances memory efficiency in replay-based continual learning, with simple methods outperforming autoencoders.

Findings

Simple compression methods outperform autoencoders in accuracy.

FETCH increases classification accuracy on CIFAR10 and CIFAR100.

Effective memory reduction in continual learning scenarios.

Abstract

Class-incremental continual learning is an important area of research, as static deep learning methods fail to adapt to changing tasks and data distributions. In previous works, promising results were achieved using replay and compressed replay techniques. In the field of regular replay, GDumb achieved outstanding results but requires a large amount of memory. This problem can be addressed by compressed replay techniques. The goal of this work is to evaluate compressed replay in the pipeline of GDumb. We propose FETCH, a two-stage compression approach. First, the samples from the continual datastream are encoded by the early layers of a pre-trained neural network. Second, the samples are compressed before being stored in the episodic memory. Following GDumb, the remaining classification head is trained from scratch using only the decompressed samples from the reply memory. We evaluate FETCH in different scenarios and show that this approach can increase accuracy on CIFAR10 and CIFAR100. In our experiments, simple compression methods (e.g., quantization of tensors) outperform deep autoencoders. In the future, FETCH could serve as a baseline for benchmarking compressed replay learning in constrained memory scenarios.