How stable are Transferability Metrics evaluations?
This addresses the problem of unreliable evaluations in transfer learning for researchers, providing a more robust framework, though it is incremental as it builds on existing metrics.
The paper tackles the inconsistency in evaluating transferability metrics by conducting a large-scale study with 715k experimental variations, revealing that small setup changes lead to different conclusions and proposing aggregation for more stable evaluations, which identifies LogME, NLEEP, and GBC as superior in specific scenarios but no single metric works best overall.
Transferability metrics is a maturing field with increasing interest, which aims at providing heuristics for selecting the most suitable source models to transfer to a given target dataset, without fine-tuning them all. However, existing works rely on custom experimental setups which differ across papers, leading to inconsistent conclusions about which transferability metrics work best. In this paper we conduct a large-scale study by systematically constructing a broad range of 715k experimental setup variations. We discover that even small variations to an experimental setup lead to different conclusions about the superiority of a transferability metric over another. Then we propose better evaluations by aggregating across many experiments, enabling to reach more stable conclusions. As a result, we reveal the superiority of LogME at selecting good source datasets to transfer from in a semantic segmentation scenario, NLEEP at selecting good source architectures in an image classification scenario, and GBC at determining which target task benefits most from a given source model. Yet, no single transferability metric works best in all scenarios.