DLS 2 W3 Batch Norm at Test Time - Why do it?

As I understand - Test time is not about training. During test time - no gradient descent, no parameter updates. We would use the trained NN (parameters) from any number of gradient descents to make predictions.
If so, then why do we care to do Batch Norm at Test Time

Hello @dds,

I think it is just why we need to normalize our test dataset in the same way we had normalized our training dataset. Your neural network’s parameters are trained to look at normalized data.

Cheers,
Raymond