SCAN: Bootstrapping Contrastive Pre-training for Data Efficiency

TL;DR

A novel dynamic bootstrapping dataset pruning method that involves pruning data preparation followed by dataset mutation operations, both of which undergo iterative and dynamic updates, and demonstrates significant superiority in terms of downstream performance over other static coreset selection approaches.

Abstract

While contrastive pre-training is widely employed, its data efficiency problem has remained relatively under-explored thus far. Existing methods often rely on static coreset selection algorithms to pre-identify important data for training. However, this static nature renders them unable to dynamically track the data usefulness throughout pre-training, leading to subpar pre-trained models. To address this challenge, our paper introduces a novel dynamic bootstrapping dataset pruning method. It involves pruning data preparation followed by dataset mutation operations, both of which undergo iterative and dynamic updates. We apply this method to two prevalent contrastive pre-training frameworks: \textbf{CLIP} and \textbf{MoCo}, representing vision-language and vision-centric domains, respectively. In particular, we individually pre-train seven CLIP models on two large-scale image-text pair datasets, and two MoCo models on the ImageNet dataset, resulting in a total of 16 pre-trained models. With a data pruning rate of 30-35\% across all 16 models, our method exhibits only marginal performance degradation (less than \textbf{1\%} on average) compared to corresponding models trained on the full dataset counterparts across various downstream datasets, and also surpasses several baselines with a large performance margin. Additionally, the byproduct from our method, \ie coresets derived from the original datasets after pre-training, also demonstrates significant superiority in terms of downstream performance over other static coreset selection approaches.

Figures (5)

• Figure 1: The interplay between training data size and model downstream performance of base model CLIP, our method SCAN, and two SoTA baselines.
• Figure 2: Overall pipeline of the proposed SCAN method. We begin by identifying a substantial portion of data samples as pruning candidates. Subsequently, a subset of these candidates is employed for pruning based on varying mutation ratios that are gradually increased (bootstrapping). After growing back to the original full dataset, the above two operations are iterated for another round.
• Figure 3: Downstream performance variation of two CLIP models w.r.t. different pruning ratios.
• Figure 4: Comparison of pre-training time between the base CLIP model and our SCAN on the CC12M+ dataset.
• Figure 5: Examples of ill-matched samples identified by our SCAN.