You are here

RBM Deep Belief Network for Visual Digit Recognition

Description: 
A spiking neuron model for digit recognition, created by training an RBM Deep Belief Network on the MNIST database, then converting the resulting model to spiking neurons via Nengo.
Keywords: 
Requirements: 
Peer Reviewed: 
No
Instructions: 

Download the attached zip file and unzip it into the Nengo directory.

Run Nengo. You may need to increase the amount of memory available to Nengo by changing the command line option -Xmx800m to -Xmx1600m (in the script nengo or nengo.bat.

Run digit.py. After a while the interactive mode display will automatically appear.

Press play to start the model running. Digits will be shown at random to the network as input (on the left). The final output (on the right) is compared to the ideal semantic pointer for each digit (lower right). The sparse spiking behaviour of the intermediate layer neurons is shown in the middle.

Figures: