batch normalization question

[ghost]

Why doesn't the batch normalization layer apply scaling and shifting like in the original paper?

thanks,
Marijke

[edgarriba]

/cc @nyanp

[nyanp]

@HotMarijke
tiny-dnn's batch normalization is ported from caffe's one, which don't have scale and bias operation. We already have scaling and shifting layer named linear_layer, and if we can train it's scale and bias, we can emulate the original paper by combine them.

[ghost]

Okay, now I understand!
Thanks for this great project!

Marijke