Visualising how my CNNs is doing?

I’m track both the loss and the accuracy of both training and test sets at training time.

However, I want to have a deeper understanding of my neural network.

  • What is the best way to inspect what the inner layers are computing ?

For example, they do it in the animation from Stanford CS site:

I’d like to do the same with my Torch based CNN.

  • Can you suggest any starting point ?

I’m also trying to start tracking the gradients in training time, and would welcome any pointer as to how could I track this in a useful, and simple way (I am rather a beginner.)