Newer is not always better: Rethinking transferability metrics, their peculiarities, stability and performance
arXiv:2110.06893 · doi:10.1007/978-3-031-26387-3_42
Abstract
Fine-tuning of large pre-trained image and language models on small customized datasets has become increasingly popular for improved prediction and efficient use of limited resources. Fine-tuning requires identification of best models to transfer-learn from and quantifying transferability prevents expensive re-training on all of the candidate models/tasks pairs. In this paper, we show that the statistical problems with covariance estimation drive the poor performance of H-score -- a common baseline for newer metrics -- and propose shrinkage-based estimator. This results in up to 80% absolute gain in H-score correlation performance, making it competitive with the state-of-the-art LogME measure. Our shrinkage-based H-score is -10 faster to compute compared to LogME. Additionally, we look into a less common setting of target (as opposed to source) task selection. We demonstrate previously overlooked problems in such settings with different number of labels, class-imbalance ratios etc. for some recent metrics e.g., NCE, LEEP that resulted in them being misrepresented as leading measures. We propose a correction and recommend measuring correlation performance against relative accuracy in such settings. We support our findings with ~164,000 (fine-tuning trials) experiments on both vision models and graph neural networks.
Accepted in ECMLPKDD 2022
References in corpus (6)
- Learning Transferable Features with Deep Adaptation Networks
- Maximum Density Divergence for Domain Adaptation
- LEEP: A New Measure to Evaluate Transferability of Learned Representations
- Rethinking the Hyperparameters for Fine-tuning
- LogME: Practical Assessment of Pre-trained Models for Transfer Learning
- A linearized framework and a new benchmark for model selection for fine-tuning