The Two Dimensions of Worst-case Training and the Integrated Effect for Out-of-domain Generalization

TL;DR

This paper introduces W2D, a new training heuristic that emphasizes worst-case scenarios across both sample and feature dimensions to improve out-of-domain generalization, validated through empirical benchmarks.

Contribution

The paper proposes a novel two-dimensional worst-case training method, merging sample and feature emphasis, to enhance model robustness and out-of-domain generalization.

Findings

W2D outperforms standard training on benchmarks.

Emphasizing both sample and feature worst-cases improves robustness.

The method is simple yet effective in practice.

Abstract

Training with an emphasis on "hard-to-learn" components of the data has been proven as an effective method to improve the generalization of machine learning models, especially in the settings where robustness (e.g., generalization across distributions) is valued. Existing literature discussing this "hard-to-learn" concept are mainly expanded either along the dimension of the samples or the dimension of the features. In this paper, we aim to introduce a simple view merging these two dimensions, leading to a new, simple yet effective, heuristic to train machine learning models by emphasizing the worst-cases on both the sample and the feature dimensions. We name our method W2D following the concept of "Worst-case along Two Dimensions". We validate the idea and demonstrate its empirical strength over standard benchmarks.