What changes in BatchNorm during inference?

Question: During inference, how does Batch Normalization usually normalize activations? A) It computes gradients through the validation set B) It uses learned running statistics instead of current mini-batch statistics C) It randomly drops neurons D) It removes all nonlinearities Correct: B Explanation: During training, BatchNorm uses mini-batch statistics and updates running estimates. During inference, it typically uses stored running mean and variance so predictions are stable and independent of batch composition. Topic: advanced ML / deep learning / batch normalization