Test-Time Adaptation with Principal Component Analysis

Thomas Cordier, Victor Bouvier, Gilles Hénaff +1

Machine Learning models are prone to fail when test data are different from training data, a situation often encountered in real applications known as distribution shift. While sti…