This is my fork of the cuda-convnet convolutional neural network implementation written by Alex Krizhevsky.

cuda-convnet has quite extensive documentation itself. Find the MAIN DOCUMENTATION HERE.

This document will only describe the small differences between cuda-convnet as hosted on Google Code and this version.

Dropout is a relatively new regularization technique for neural networks. See the Improving neural networks by preventing co-adaptation of feature detectors and Improving Neural Networks with Dropout papers for details.

To set a dropout rate for one of our layers, we use the dropout parameter in our model's layer-params configuration file. For example, we could use dropout for the last layer in the CIFAR example by modifying the section for the fc10 layer to look like so:

[fc10]
epsW=0.001
epsB=0.002
# ...
dropout=0.5

In practice, you'll probably also want to double the number of outputs in that layer.

An environment variable CONVNET_RANDOM_SEED, if set, will be used to set the CURAND library's random seed. This is important in order to get reproducable results.

The build configuration and code has been updated to work with CUDA 5.5. The paths in build.sh and Makefile were updated to work with the default install directories of the CUDA Toolkit 5.5.