How often does dropout get updated

Hi guys.

I would like to know, if Dropout (dropped out nodes) gets updated once on every input or once every batch.

Asked this question on stackOverFlow aswell, if you are interested in answering there:

Thanks!

Check the docstring:

https://github.com/keras-team/keras/blob/master/keras/layers/core/dropout.py#L29:L32

1 Like

I think this question is still valid, because the dropout being updated per batch does not necessarily mean that the same weights are dropped out per train input. I want to know if I can have the same input, slightly augmented, in one batch, or if they need to always be seperated. Thanks!