This project's aim is to demonstrate batch normalization using handwritten digits recognition task as an example. Relevant documentation is in the tutorial about neural networks.
How To Build
$ wget -qO- https://get.haskellstack.org/ | sh
curl -sSL https://get.haskellstack.org/ | sh)
Build and run MNIST benchmark on all available cores:
10 training epochs Training accuracy (SGD + batchnorm) 100.0 Validation accuracy (SGD + batchnorm) 98.1 Training accuracy (SGD) 85.7 Validation accuracy (SGD) 85.8