Below, are my responses to the 5 exercises at the bottom of Lab 4 - Using Convolutions!
As the number of filters increase in the convolution layer, from 16 to 64, run time increases dramtically, and accuracy and validation increase.
Run time was about the same, but accuracy decreased a little.
I added 1 more layer, and it took much longer to run. Accuracy increased from .9866 to .9907.
Accuracy decreased.
I was able to implement it! It cancelled training after 4 epochs.