robust-attribution-regularization icon indicating copy to clipboard operation
robust-attribution-regularization copied to clipboard

Published 20 hours ago • verified publisher icon jfc43

Getting nan loss on CIFAR10 dataset

Open msingh27 opened this issue 5 years ago • 1 comments

Hi,

Can you please suggest modifications from GTSRB's config (specially learning rate schedule) so that this code runs on CIFAR10 dataset?

Sample output: Step 100: (2019-09-27 10:58:14.473903) training nat accuracy 7.0%, loss nan training adv accuracy 7.0%, loss nan training IG term nan

msingh27 avatar Sep 27 '19 11:09 msingh27

Hi,

I think you need to try smaller learning rate.

jfc43 avatar Oct 10 '19 21:10 jfc43