ConvNetJS is a browser-based deep learning library that implements neural network modules, convolutional layers, classification and regression cost functions. It includes an experimental reinforcement learning module based on Deep Q Learning with interactive browser demos for MNIST, CIFAR-10, and other datasets.