0
$\begingroup$

I am novice in Neural Networks and particularly with its implementation in Mathematica. I am trying to understand the graphical result of NetGraph function but it results in non-classical displaying of ANN - representation is linear, but I expect to see layers and neurons, like on this image.

How should I interpret the result of NetGraph visual output and how to build e.g. XOR neural network as on the image above?

$\endgroup$
  • $\begingroup$ The idea as I understand it is to imitate what tensorflow does: tensorflow.org/programmers_guide/graphs $\endgroup$ – b3m2a1 Apr 2 '18 at 16:15
  • $\begingroup$ Modern deep neural networks often have millions of neurons and weights (e.g. for each pixel in an image) - so it isn't practical to draw each neuron individually. Instead, you draw layers of neurons that perform the same function, like dot product or convolution $\endgroup$ – Niki Estner Apr 3 '18 at 14:14
1
$\begingroup$

I am trying to understand the graphical result of NetGraph function but it results in non-classical displaying of ANN

The image you sent is not of a NetGraph, but rather a chain: NetChain[{LinearLayer[2], LinearLayer[1]}]. It would be simple to write your own visualizer for this similar to the image you sent, but we don't have built-in functionality for this. The reason: its super-specific to the LinearLayer (how do convolutions look? or attention layers? or...). Basically, this style of graph is only interesting for tiny toy nets composed of LinearLayers.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.