Target Robust Discriminant Analysis

TL;DR

This paper introduces robust parameter estimators for discriminant analysis that improve adaptive classifier performance across varying domain shifts, addressing limitations of existing domain adaptation methods.

Contribution

It proposes a novel robust discriminant analysis method that guarantees performance gains over non-adaptive classifiers under domain shift.

Findings

Robust estimators improve classification accuracy under domain shift.

The method outperforms traditional domain adaptation approaches.

Guarantees performance improvement over non-adaptive classifiers.

Abstract

In practice, the data distribution at test time often differs, to a smaller or larger extent, from that of the original training data. Consequentially, the so-called source classifier, trained on the available labelled data, deteriorates on the test, or target, data. Domain adaptive classifiers aim to combat this problem, but typically assume some particular form of domain shift. Most are not robust to violations of domain shift assumptions and may even perform worse than their non-adaptive counterparts. We construct robust parameter estimators for discriminant analysis that guarantee performance improvements of the adaptive classifier over the non-adaptive source classifier.