search

zqs1022 / interpretableCNN

MIT License
223 stars 52 forks source link

Technique mistake in the paper #2

Closed hangg7 closed 5 years ago

hangg7 commented 6 years ago

Hi,

About computing p(x|T), the technique currently listed in the paper is wrong:

It's clear that you are trying to normalize all scores in a batch using softmax where score function is the sum over masked activations, i.e.,

image

however the trace operation is not correct, since it would end in

image

Please consider the modification - it might be a minor mistake however resulting in unnecessary confusion.

xizero00 commented 5 years ago

Hi @cullengao , I also find this typo.