Joakim Skarding

An MNIST classifier that applies a simple neural network to classify handwritten digits. The MNIST data set needed to run this classifier is automatically downloaded. The Lasagne tutorial served as significant inspiration.

As the code stands, the classifier runs with the following:
• One hidden layer
• 625 hidden units
• A batch size of 100
• 100 epochs

One run before uploading yielded an accuracy of 97.29% on the test set.

The main features include:
• Loading the image files from yann.lecun.com/exdb/mnist/.
• Displaying ten randomly selected digits together with corresponding labels before classification.
• Running the classification with a batch size of 100.
• Visualizing average training and test loss.
• Displaying ten randomly selected digits together with the predictions from the network.
• Visualizing the weight matrices of ten arbitrary units.

Spent a year as an exchange student studying Computer Engineering at the prestigious Seoul National University in South Korea from 2014 to 2015.