Selection via Proxy: Efficient Data Selection for Deep Learning

TL;DR

This paper introduces a proxy-based data selection method that significantly accelerates active learning and core-set selection in deep learning by using small, fast-to-train models as proxies, maintaining accuracy while reducing computation.

Contribution

The authors propose 'selection via proxy' (SVP), a novel approach that employs lightweight proxy models for efficient data selection in deep learning tasks, reducing training time substantially.

Findings

SVP achieves an order of magnitude faster data selection in active learning.

Proxies over 10x faster can remove up to 50% of data without accuracy loss.

End-to-end training time improves by up to 1.6x with SVP.

Abstract

Data selection methods, such as active learning and core-set selection, are useful tools for machine learning on large datasets. However, they can be prohibitively expensive to apply in deep learning because they depend on feature representations that need to be learned. In this work, we show that we can greatly improve the computational efficiency by using a small proxy model to perform data selection (e.g., selecting data points to label for active learning). By removing hidden layers from the target model, using smaller architectures, and training for fewer epochs, we create proxies that are an order of magnitude faster to train. Although these small proxy models have higher error rates, we find that they empirically provide useful signals for data selection. We evaluate this "selection via proxy" (SVP) approach on several data selection tasks across five datasets: CIFAR10, CIFAR100, ImageNet, Amazon Review Polarity, and Amazon Review Full. For active learning, applying SVP can give an order of magnitude improvement in data selection runtime (i.e., the time it takes to repeatedly train and select points) without significantly increasing the final error (often within 0.1%). For core-set selection on CIFAR10, proxies that are over 10x faster to train than their larger, more accurate targets can remove up to 50% of the data without harming the final accuracy of the target, leading to a 1.6x end-to-end training time improvement.