Revisiting the Robustness of the Minimum Error Entropy Criterion: A Transfer Learning Case Study

TL;DR

This paper investigates the robustness of the minimum error entropy (MEE) criterion in transfer learning, demonstrating its effectiveness against covariate shift and showing it can improve existing algorithms in real-world tasks.

Contribution

It provides a new theoretical proof of MEE's robustness to covariate shift and shows that replacing MSE with MEE enhances transfer learning performance.

Findings

MEE is robust against covariate shift in transfer learning.

Replacing MSE with MEE yields competitive results on real-world data.

Theoretical analysis supports MEE's effectiveness in noisy, real-life scenarios.

Abstract

Coping with distributional shifts is an important part of transfer learning methods in order to perform well in real-life tasks. However, most of the existing approaches in this area either focus on an ideal scenario in which the data does not contain noises or employ a complicated training paradigm or model design to deal with distributional shifts. In this paper, we revisit the robustness of the minimum error entropy (MEE) criterion, a widely used objective in statistical signal processing to deal with non-Gaussian noises, and investigate its feasibility and usefulness in real-life transfer learning regression tasks, where distributional shifts are common. Specifically, we put forward a new theoretical result showing the robustness of MEE against covariate shift. We also show that by simply replacing the mean squared error (MSE) loss with the MEE on basic transfer learning algorithms such as fine-tuning and linear probing, we can achieve competitive performance with respect to state-of-the-art transfer learning algorithms. We justify our arguments on both synthetic data and 5 real-world time-series data.