What is covariate shift?

Question: In machine learning, what does covariate shift mean? A) The conditional label distribution P(y|x) changes, but P(x) remains fixed B) The input feature distribution P(x) changes between training and testing, while P(y|x) is assumed stable C) The model loss function changes during inference D) The number of classes increases after deployment Correct: B Explanation: Covariate shift occurs when the distribution of input features changes between training and deployment, but the relationship between features and labels is assumed to remain mostly the same. This can hurt generalization because the model sees test inputs from regions not well represented during training. Topic: advanced ML / distribution shift / generalization