Core Tokensets for Data-efficient Sequential Training of Transformers

TL;DR

This paper introduces core tokensets, a novel method for data-efficient sequential training of transformers that selects the most informative tokens and features, significantly reducing memory while maintaining high performance across various tasks.

Contribution

The paper proposes core tokensets, a new approach that constructs token-level summaries for efficient continual learning, surpassing traditional sample-based coresets in performance and memory efficiency.

Findings

Core tokensets achieve comparable performance with only 1% data.

Significant memory reduction in incremental learning tasks.

Effective across image classification, visual QA, and captioning.

Abstract

Deep networks are frequently tuned to novel tasks and continue learning from ongoing data streams. Such sequential training requires consolidation of new and past information, a challenge predominantly addressed by retaining the most important data points - formally known as coresets. Traditionally, these coresets consist of entire samples, such as images or sentences. However, recent transformer architectures operate on tokens, leading to the famous assertion that an image is worth 16x16 words. Intuitively, not all of these tokens are equally informative or memorable. Going beyond coresets, we thus propose to construct a deeper-level data summary on the level of tokens. Our respectively named core tokensets both select the most informative data points and leverage feature attribution to store only their most relevant features. We demonstrate that core tokensets yield significant performance retention in incremental image classification, open-ended visual question answering, and continual image captioning with significantly reduced memory. In fact, we empirically find that a core tokenset of 1\% of the data performs comparably to at least a twice as large and up to 10 times larger coreset.