1. Run CNNTF.py and print out the shapes of the following variables from conv_net (after each assignment of that variable): x, conv1, conv2, fc1, out.

   Do the same after we replace the size of filters in the second convolutional layer (for conv2) from 3 to 5.

   What is the impact to the accuracy when the filter size is changed from 3 to 5? Look for best learning rates for the changed filter and report the experimental results.

2. Various activation functions are provided in Tensorflow: https://www.tensorflow.org/api_guides/python/nn Please experiment these three activation functions in the convolutional and dense layers in conv_net: tf.nn.relu, tf.nn.sigmod, and tf.nn.tanh. To simplify the experiment, for one experiment, one activation function is used for all the layers. Look for best learning rates for each activation function and report the experimental results.

Please submit everything required, including the changed code and output of a sample run, in the ICON dropbox for Assignment 5 before the deadline.

Thank you!