What is convnetjs?

ConvNetJS is a JavaScript library for building and training neural networks, the kind of software that learns patterns from data. What makes it unusual is that it runs entirely in a web browser, with no server or installation required. You can train a network on image data, run it on new inputs, and watch the results, all on a webpage. The library supports the main building blocks used in modern machine learning: standard fully connected layers, convolutional layers (which are specialized for processing images), and an experimental module for reinforcement learning, where a model learns by trial and error rather than from labeled examples. Classification (sorting inputs into categories) and regression (predicting numeric values) are both supported. The README includes a worked example showing how to define a small network in a few lines of JavaScript, train it on a data point, and see the probability output shift after training. There are also online demos linked in the README covering handwritten digit recognition, image classification on a standard benchmark dataset, and a reinforcement learning agent. The library can be loaded directly into a webpage, or installed via npm for use in a Node.js environment. A prebuilt and minified version is available for download. If you want to modify the source, a compile step using the Ant build tool is needed to combine the source files and produce the final library file. The author notes in the README that the project is no longer actively maintained and that the npm package may not work. The source code and demos remain available for learning and reference purposes. The license is MIT.