Part 2 wish list: neural net debugging

If I could have a wish for the 2nd part it would be:

  1. NN debugging. Jeremy mentioned in lesson 5 (around 1:44) that it can be very helpful if we knew how to debug NNs; get the values of gradients and know how to plot them, and e.g. assess whether they are getting worse with each layer.

I’d also be interested in hearing about YOLO and auto encoders but just NN debugging would help me tremendously. Thank you!


This is a very useful topic. I would love to know what tools etc. are suggested.

A quick google search shows people are using Tensorboard.

1 Like

Check out the Pytorch forum (separate login required) and search for “Using Tensorboard with Pytorch?” for more information. I haven’t tried myself but it looks promising.