A SAS DLPy model uses Batch Normalization (BN) layers between Conv2d layers. How do Batch Normalization layers behave differently during inference (scoring) compared to during training?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Here's the deal — b is correct because Batch Normalization has two distinct operational modes. During training, it normalizes using the current mini-batch's mean and variance (which vary per batch), reducing internal covariate shift.
Full explanation below image
Full Explanation
B is correct because Batch Normalization has two distinct operational modes. During training, it normalizes using the current mini-batch's mean and variance (which vary per batch), reducing internal covariate shift. During inference, using single-observation or small-batch statistics would be noisy, so BN instead uses exponential moving averages of mean and variance accumulated across all training batches. These running statistics are fixed during inference. A is wrong; BN does not apply dropout. C is wrong; BN layers remain in the inference graph. D is wrong; training uses batch statistics, not global statistics.