Finding the Needle in the Pre-Trained Model Zoo

Finding the Needle in the Pre-Trained Model Zoo: The Use of Rich Metadata and Graph Learning to Estimate Task Transferability

van der Wilk, Hilco (TU Delft Electrical Engineering, Mathematics and Computer Science)

Hai, R. (mentor)
Li, Z. (mentor)
Anand, A. (graduation committee)
Song, Q. (graduation committee)

Delft University of Technology

Computer Science

2024-06-25

The democratization of machine learning through public repositories, often known as model zoos, has significantly increased the availability of pre-trained models for practitioners. However, this abundance can make it difficult to choose the most suitable pre-trained model for fine-tuning on new tasks. Although various methods have been proposed in the field of transferability estimation to address this issue, these methods can take hours to execute and may still fail to find the optimal pre-trained model for fine-tuning. By exploring a new graph learning-based approach to transferability estimation, we outperform state-of-the-art methods such as LogME, improving the accuracy of the best-predicted model by up to 31.5\% in less than 5 minutes.

transferability estimation
transfer learning
deep learning
model zoos
graph learning

http://resolver.tudelft.nl/uuid:59a7191d-4522-4a74-8ccf-503d11a5101b

Student theses

master thesis

© 2024 Hilco van der Wilk