Imaginghub Forum - How can I build custom CNN with cuDNN?

I would like to test a modified back-propagation algorithm for a custom CNN, but some existed frameworks such as TensorFlow didn't allow me to do it. Maybe that's because my bp algorithm is not the same as the TF's, so I guess, I need to implement the algorithm from scratch using cuDNN. But where can I start? Are there any simple samples or tips?

Answers

Olya_Olya, actually, it`s not necessary to work on such a low level. Tensorflow allows overwriting the backpropagation algorithm by your own approach. You can see some good examples here:https://medium.com/datadriveninvestor/override-tensorflow-backward-propagation-1a7e2bb3f516

Thanks a lot! And what would you recommend besides TensorFlow?

I guess, if you want to use your own CNN, you can try to implement it from scratch with pandas and numpy.

Add Answer

How can I build custom CNN with cuDNN?

How can I build custom CNN with cuDNN?

Answers

Need support?