GRASP: A Rehearsal Policy for Efficient Online Continual Learning

TL;DR

GRASP is a novel rehearsal policy for continual learning that efficiently selects prototypical samples, improving accuracy and reducing updates across multiple datasets with minimal overhead.

Contribution

It introduces GRASP, a new sample selection policy that outperforms existing methods in continual learning with minimal additional computational cost.

Findings

GRASP achieves higher accuracy than 17 other policies on ImageNet.

GRASP reduces the number of updates by 40% while maintaining performance.

Effective for both image and text classification tasks.

Abstract

Continual learning (CL) in deep neural networks (DNNs) involves incrementally accumulating knowledge in a DNN from a growing data stream. A major challenge in CL is that non-stationary data streams cause catastrophic forgetting of previously learned abilities. A popular solution is rehearsal: storing past observations in a buffer and then sampling the buffer to update the DNN. Uniform sampling in a class-balanced manner is highly effective, and better sample selection policies have been elusive. Here, we propose a new sample selection policy called GRASP that selects the most prototypical (easy) samples first and then gradually selects less prototypical (harder) examples. GRASP has little additional compute or memory overhead compared to uniform selection, enabling it to scale to large datasets. Compared to 17 other rehearsal policies, GRASP achieves higher accuracy in CL experiments on ImageNet. Compared to uniform balanced sampling, GRASP achieves the same performance with 40% fewer updates. We also show that GRASP is effective for CL on five text classification datasets.