Train Deep Learning

ConvNetJS is a javascript implementation of deep learning created by Andrej Karpathy here here. Intro → Browser Demos section so you can jump straight into hands-on neural-network experiments inside the browser. Each tile opens the vendored local copy of the original Stanford demos in a new tab—no installs or external network dependency needed.

Inverted Pendulum RL

Train a cart-pole balancer with an in-browser CEM agent and watch the pole stabilize live.

Lunar Lander PG

Run REINFORCE on a 2-D lunar lander with live returns and best-policy playback.

MNIST CNN

Train a convolutional net that recognizes handwritten digits.

CIFAR-10 CNN

Classify natural images from CIFAR-10 with a deeper convolutional net.

2-D Classifier

Draw toy datasets and watch a small net learn decision boundaries live.

1-D Regression

Interactively fit smooth curves to 1-D data using neural nets.

MNIST Autoencoder

Compress and reconstruct digits with an autoencoder architecture.

Deep Q Learning

Explore reinforcement learning agents using Deep Q-Networks.

Neural Painter

Watch a fully-connected net “paint” an image pixel by pixel.

Optimizer Playground

Compare SGD, Adagrad, and Adadelta on the same dataset in real time.

Want to dive deeper? Browse the full ConvNetJS intro page or read the detailed documentation.