A Data-Based Perspective on Transfer Learning

TL;DR

This paper investigates how the composition of source datasets affects transfer learning performance, introducing a framework to identify and remove detrimental data points, thereby improving downstream task results.

Contribution

It presents a novel framework for analyzing source dataset impact in transfer learning, enabling detection of data issues and enhancing transfer performance.

Findings

Removing detrimental data points improves transfer results

Framework can identify data leakage and misleading examples

Source dataset composition significantly influences transfer learning success

Abstract

It is commonly believed that in transfer learning including more pre-training data translates into better performance. However, recent evidence suggests that removing data from the source dataset can actually help too. In this work, we take a closer look at the role of the source dataset's composition in transfer learning and present a framework for probing its impact on downstream performance. Our framework gives rise to new capabilities such as pinpointing transfer learning brittleness as well as detecting pathologies such as data-leakage and the presence of misleading examples in the source dataset. In particular, we demonstrate that removing detrimental datapoints identified by our framework improves transfer learning performance from ImageNet on a variety of target tasks. Code is available at https://github.com/MadryLab/data-transfer