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 increase the number of
hidden units in that layer (outputs), and change the wc and
initW parameters, too.
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.
Minor changes include using more Debian/Ubuntu default for the paths
defined in the build.sh script and Makefile.