Gradient for loss function

[waynezw0618]

Dear Sir
I found your paper which is very interesting and useful. and I am trying to implement the dynamic coefficient according to the first method you have introduced. I am wondering in practice how to implement the gradient of the loss. I have tried auto differentiation like:
tf.gradients(loss,weight)
But I don't know whether it is the one for your definition.

thanks for replying

[RahulSundar]

Dear Dr. Paris Perdikaris

I am Rahul Sundar, a research scholar from Indian Institute of Technology. I found your group's work really intriguing as it allies with my area of interest currently.
I have been presently trying to work on the "Gradient pathologies" paper. As discussed by Mr. Wei Zhang, I too am facing the same confusion with regard to the gradients of loss function at specific layers. Would like to get some clarification on the math if possible.

Thanks for replying