Improving on a single-layer convolutional architecture with lateral regularization and 
allowing breaking convolutional simmetry.

---

Working example of an autoencoder learning features on the MNIST dataset will run with

python autoencoder.py

Working example (sort of) of a conv_net with autoencoder pretraining with a 10x10 receptive field
will run with

python conv_net.py

Requires skdata, numpy and theano

---

Alex Susemihl 2013/07/02