Thank you :)

I did look at some of your content. I especially liked the triplet loss piece. I also agree it is a great technique that is not often talked about.

Out of all crazy ideas I have read about in the past year, training only batch norm layers is definitively one of the craziest. I think you would also agree.