Live JavaScript-based in-browser demo of "Attend, Infer, Repeat" - unsupervised scene understanding model by Eslami et al., 2016. Draw digits in the left canvas, clear with a right click. On startup the demo downloads ~5MB of model parameters.
Implemented using deeplearn.js, with the trained model parameters values obtained from this implementation of AIR. This is a work in progress, some GUI and additional features are yet to be added.