Fork me on GitHub
ConvNetJS is a Javascript library for training Deep Learning models (mainly Neural Networks) entirely in your browser. Open a tab and you're training. No software requirements, no compilers, no installations, no GPUs, no sweat.

Browser Demos

Description

The library allows you to formulate and solve Neural Networks. Current support includes:

Code

The code is available on Github under MIT license and I warmly welcome pull requests for new features / layers / demos and miscellaneous improvements. The library is also available on npm for use in Nodejs, under name convnetjs.

About

This project was started by @karpathy. I am a PhD student at Stanford studying Machine Learning and Computer Vision and I've worked on Deep Learning both as part of my research and as an intern at Google (multiple times). In early versions of this code I chose to go first for simplicity, core concepts and most common use cases, though many bells and whistles can be added over time to add modeling flexibility and improve training times.

In general, people in the field are eager to make neural nets bigger and faster so Javascript may seem like a strange choice, but I do believe that there are multiple interesting use cases:

The downside, of course, is that we're paying about an order of magnitude of efficiency in speed compared to native implementations, but once I get around implementing the core operations in WebGL this could be effectively eliminated.

I elaborate in much more detail on the back story behind the project, my motivations and potential uses in this interview with Data Science Weekly.

Discussion Group