Target Robust Discriminant Analysis

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