JohnCalhoun commented on issue #10002: General support of OPs for second-order 
gradient
URL: 
https://github.com/apache/incubator-mxnet/issues/10002#issuecomment-429563546
 
 
   @aidan-plenert-macdonald so I have been looking in to how the unary ops 
work. I don't think it is possible to register compound ops. It seems the 
engine is not doing arbitrary differentiation, it just looks up what was 
registered as the derivative. so -sin(x) has to be its own op and there is no 
way to describe it is mul(-1,sin(x)) with mul and sin being ops. Am I 
understanding this right?
   
   sin(x)->cos(x) is simple but I think that is a corner case. 
   
   It might now be possible to support arbitrary order gradients with out 
writing an endless amount of _backward_* ops. Writing enough ops to support 
second order gradients is for sure possible. 

----------------------------------------------------------------
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

Reply via email to