For the National Meeting of Data-Intensive Science CDTs in Edinburgh.
Train some simple fully-connected and convolutional networks on the MNIST dataset, with data augmentation to squeeze out a few more percent accuracy.
MNIST_Edinburgh.ipynb - bare version
MNIST_Edinburgh_presentation.ipynb - version with outputs, extra material and figures
These Jupyter notebooks are set up for presenting with RISE.
If GitHub is not rendering the notebooks, you can view them at nbviewer: http://nbviewer.jupyter.org/github/bamford/keras_demo/