Modeling Binary and Unary Operators with an Artifical Neural Network

To develop an understanding of tensorflow’s ANN I implemented regression models of several unary and binary functions.

I created a Jupyter notebook for modeling unary functions available here or viewed online at GitHub. And here is a notebook for binary functions or online.

I show examples of 1 and 2 hidden layer ANNs. In a single hidden layer, I show that unary function are sums of scaled ReLus, where the number of ReLus is the number of nodes in the layer. With 2 layers, I show that it is the sum of piecewise linear functions, where the number of pieces is the number of nodes in the first layer.

This provides some insight to how ANN work on simple problems. Note that these notebooks use tf.contrib.learn.DNNRegressor to do a regression, not a classification.