Create a subset of the MNIST dataset:

In[4]:=

Out[4]=

Remove the last linear layer of the net, which will be used as a feature extractor:

In[5]:=

Out[5]=

Visualize the features of a subset of the MNIST dataset:

In[6]:=

Out[6]=

Extract the image reconstruction part:

In[7]:=

Out[7]=

Extract the DigitCaps feature vector for a given digit image:

In[8]:=

Out[8]=

Reconstruct the image from the feature vector:

In[9]:=

Out[9]=

Experiment with changing the feature vector. Add a shift along a single coordinate at a time:

In[10]:=

Out[10]=

In[11]:=

Out[11]=

In[12]:=

Out[12]=

Retrieve the uninitialized training architecture:

In[13]:=

Out[13]=

Retrieve the MNIST dataset:

In[14]:=

Out[14]=

Use the training dataset provided:

In[15]:=

Out[15]=

Use the test dataset provided:

In[16]:=

Out[16]=

Initialize the “W” matrices properly:

In[17]:=

Out[17]=

Train the net (if a GPU is available, setting TargetDevice -> "GPU" is recommended):

In[18]:=

Out[18]=

Inspect the number of parameters of all arrays in the net:

In[19]:=

Out[19]=

Obtain the total number of parameters:

In[20]:=

Out[20]=

Obtain the layer type counts:

In[21]:=

Out[21]=

Display the summary graphic:

In[22]:=

Out[22]=

Export the net into a format that can be opened in MXNet:

In[23]:=

Out[23]=

Export also creates a net.params file containing parameters:

In[24]:=

Out[24]=

Get the size of the parameter file:

In[25]:=

Out[25]=

The size is similar to the byte count of the resource object:

In[26]:=

Out[26]=

Represent the MXNet net as a graph:

In[27]:=

Out[27]=