Nicolas Thiébaud home

Visualizing the training of a logistic regression built with theano

Tuesday, September 3 2013 – San Francisco

This work is simply the output of the logisitic regression tutorial.

The tutorial walks the user through the conception of a two layer neural network. The first layer acts as the input vector and the output layer are the features we are trying to detect (ie: the actual number).

The dataset used is the MNIST database of handwritten digits. This is a visualisation of the weight matrix and bias vector during the training.

All the code to create this graph can be found here. Checkout the Makefile.