Repository files navigation

GPU accelerated handwritten digit recognition withregl.

Demo here

This demo does handwritten digit recognition by evaluating aConvolutional Neural Network on the GPU with WebGL. The network wastrained in TensorFlowby this script, and the network was thenreimplemented on the GPU by hand with WebGL. The main purpose of thedemo was to demonstate how our WebGL frameworkregl can be used to greatlysimplify GPGPU programming in WebGL. The secondary purpose was totest whether evaluating Deep Learning networks in WebGL is doable. Toour knowledge, our implementation is the first implementation ever toattempt GPU accelerating neural networks with WebGL And we hope thatthis implementation will provide a foundation for people who, like us,wish to experiment with Deep Learning and WebGL The GPU implementationcan be foundhere

Note that this network will probably be slower than the correspondingnetwork implemented on the CPU. This is because of the overheadassociated with transferring data to and from the GPU. But in thefuture we will attempt implementing more complex networks in the browser,such asNeural Style, and then we think that we will see asignificant speedup compared to the CPU.

npm install

To then run the demo, do

npm run start

To run the test cases, do

npm runtest