Exemplar-free Continual Learning of Vision Transformers via Gated Class-Attention and Cascaded Feature Drift Compensation

TL;DR

This paper introduces a novel exemplar-free continual learning method for Vision Transformers that uses gated class-attention and feature drift compensation to balance learning new tasks and retaining previous knowledge without needing past samples.

Contribution

It proposes gated class-attention and cascaded feature drift compensation techniques for exemplar-free ViT continual learning, avoiding the need for rehearsal samples.

Findings

Achieves competitive performance on CIFAR-100, Tiny-ImageNet, and ImageNet100.

Effectively limits catastrophic forgetting without exemplar replay.

Maintains task-agnostic inference capability.

Abstract

We propose a new method for exemplar-free class incremental training of ViTs. The main challenge of exemplar-free continual learning is maintaining plasticity of the learner without causing catastrophic forgetting of previously learned tasks. This is often achieved via exemplar replay which can help recalibrate previous task classifiers to the feature drift which occurs when learning new tasks. Exemplar replay, however, comes at the cost of retaining samples from previous tasks which for many applications may not be possible. To address the problem of continual ViT training, we first propose gated class-attention to minimize the drift in the final ViT transformer block. This mask-based gating is applied to class-attention mechanism of the last transformer block and strongly regulates the weights crucial for previous tasks. Importantly, gated class-attention does not require the task-ID during inference, which distinguishes it from other parameter isolation methods. Secondly, we propose a new method of feature drift compensation that accommodates feature drift in the backbone when learning new tasks. The combination of gated class-attention and cascaded feature drift compensation allows for plasticity towards new tasks while limiting forgetting of previous ones. Extensive experiments performed on CIFAR-100, Tiny-ImageNet and ImageNet100 demonstrate that our exemplar-free method obtains competitive results when compared to rehearsal based ViT methods.