Which Model to Transfer? A Survey on Transferability Estimation
This survey addresses the problem of efficiently selecting pre-trained models for transfer learning, which is critical for researchers and practitioners dealing with large-scale model availability, but it is incremental as it organizes existing work rather than introducing new methods.
The paper surveys model transferability estimation, which aims to quantify the suitability of pre-trained models for target tasks without training them individually, addressing the computational challenge of evaluating numerous models. It reviews existing advances, categorizes them into source-free and source-dependent realms, and provides a taxonomy and future directions.
Transfer learning methods endeavor to leverage relevant knowledge from existing source pre-trained models or datasets to solve downstream target tasks. With the increase in the scale and quantity of available pre-trained models nowadays, it becomes critical to assess in advance whether they are suitable for a specific target task. Model transferability estimation is an emerging and growing area of interest, aiming to propose a metric to quantify this suitability without training them individually, which is computationally prohibitive. Despite extensive recent advances already devoted to this area, they have custom terminological definitions and experimental settings. In this survey, we present the first review of existing advances in this area and categorize them into two separate realms: source-free model transferability estimation and source-dependent model transferability estimation. Each category is systematically defined, accompanied by a comprehensive taxonomy. Besides, we address challenges and outline future research directions, intending to provide a comprehensive guide to aid researchers and practitioners.