Heavy Labels Out! Dataset Distillation with Label Space Lightening

TL;DR

This paper introduces HeLlO, a label-lightening framework for dataset distillation that generates synthetic labels online, drastically reducing storage needs while maintaining high performance on large-scale datasets.

Contribution

The paper proposes a novel label-lightening approach using open-source foundation models and low-rank fine-tuning to generate synthetic labels efficiently during dataset distillation.

Findings

Achieves comparable performance with only 0.003% storage of original soft labels.

Uses foundation models like CLIP for label generation.

Demonstrates effectiveness on large-scale datasets.

Abstract

Dataset distillation or condensation aims to condense a large-scale training dataset into a much smaller synthetic one such that the training performance of distilled and original sets on neural networks are similar. Although the number of training samples can be reduced substantially, current state-of-the-art methods heavily rely on enormous soft labels to achieve satisfactory performance. As a result, the required storage can be comparable even to original datasets, especially for large-scale ones. To solve this problem, instead of storing these heavy labels, we propose a novel label-lightening framework termed HeLlO aiming at effective image-to-label projectors, with which synthetic labels can be directly generated online from synthetic images. Specifically, to construct such projectors, we leverage prior knowledge in open-source foundation models, e.g., CLIP, and introduce a LoRA-like fine-tuning strategy to mitigate the gap between pre-trained and target distributions, so that original models for soft-label generation can be distilled into a group of low-rank matrices. Moreover, an effective image optimization method is proposed to further mitigate the potential error between the original and distilled label generators. Extensive experiments demonstrate that with only about 0.003% of the original storage required for a complete set of soft labels, we achieve comparable performance to current state-of-the-art dataset distillation methods on large-scale datasets. Our code will be available.