Identifying the number of hidden nodes

[WWD]

Course: AI and Machine Learning Professional

Module: Natural Language and Computer Vision

Topic: Intro to Deep Learning

Time Point in the Video: 0:51 

In this video, the presenter is building a CNN.  On line 23 of the demonstration code we have the following line:

  AddLayer / model = BatchDLL name = 'Batchlayer2' layer = {Batchnorm act = 'Tanh} srclayer = {'HLayer2}

How many nodes are in this hidden layer?  In the previous line of code within the demo, the user had to explicitly state the number of nodes.  In the code example contained within the message, there is no mention of the number of nodes.

If it is possible, may I get a picture of the network?  In the module for CNNs, the presenter prints out pictures of the CNN.

Thank you,

Bill Donaldson

[JasonColon]

Hi Bill,

In this layer we don't specify nodes because it is a batch normalization layer. The layer that's being normalized is HLayer2, which has 20 hidden units, so we're taking the 20 hidden unit outputs and normalizing them prior to applying our non-linear activation function. We then take these 20 outputs and send them to the next HLayer which has 10 hidden units. I've attached an image below of the NN we end up building in this demo. What isn't explicitly shown is the batch normalization. You can think of that normalization occurring when we're connecting the lines from one hidden layer to the next.

Thanks,

Jason