Adding a small readme

Adding paper explaining what have I done here

Adding model fully connected with two layers (hidden layer of size 900 and tanh activation)

Adding fully connected network, to test dataset on a simple model

Calculating validation accuracy every intermediate number of steps

Adding new DNN, mobilenet (using tf-slim implementation)

Separating validation from testing

Adjusting LeCun's version (no dropout was used in the original)

Modifying nets to accept an arbitrary type of activation

Moving `model_fn' from duplicated code to principal file

Implementing Net proposed in LeCun's paper

Fixing shapes of input tensors

Constructing model from raw tensorflow values

Passing parameters needed for convolutional net

Refactoring, moving conv model to its own file