haojin2 commented on issue #11544: [MXNET-620][WIP][Do not merge]Fix flaky test batchnorm training URL: https://github.com/apache/incubator-mxnet/pull/11544#issuecomment-402625488 @marcoabreu if you take a closer look at "finding 3", we do deliver the same results as the CuDNN version, so the root cause is just that the numeric gradient having a larger error margin compared to the symbolic backward one.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
