Sukiyaki JavaScript Library

The Fastest Deep Learning Library for JavaScript

Sukiyaki is being developed as the fastest Deep Learning Library for JavaScript. It uses Sushi ( https://github.com/mil-tokyo/sushi ) inside for the fast matrix calculation.

Related papers are available ( http://mil-tokyo.github.io/miljs.html ).

Technical Features

Support of GPGPU and multi-core CPU

You can use GPGPU and multi-core CPU via Sushi. If your system does not support WebCL, Sukiyaki uses standard JavaScript automatically.

Support of Deep Convolutional Neural Network

How to Try

initialize and download dataset

node install
cd ./sample/dataset/mnist
# or
cd ./sample/dataset/cifar
./download.sh

If you don't have node,

mkdir node_modules
git clone https://github.com/mil-tokyo/sushi node_modules/milsushi

try sample program with node.js

node ./sample/node/main
> mnist

try sample program with browsers

./sample/browser/server.sh
and access http://localhost:64649/sample/browser

If you would like to use GPGPU, check Sushi ( https://github.com/mil-tokyo/sushi ) and install a WebCL implementation.