FETCH: A Memory-Efficient Replay Approach for Continual Learning in Image Classification
This work addresses memory constraints in continual learning for image classification, offering an incremental improvement over existing replay techniques.
The paper tackles the memory inefficiency of replay-based continual learning by proposing FETCH, a two-stage compression method that encodes samples with a pre-trained network and compresses them before storage, achieving increased accuracy on CIFAR10 and CIFAR100 datasets.
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.