DATE
29 October 2018 10:25 AM
The talk is a part of series of talks by highly accomplished Machine Learning Researchers from University of Montreal, Google and Google Brain for the course Deep Learning and Related Methods for Large Dataset Information Processing taught by Michal Fabinger at University of Tokyo.
In his most recent work World Models David Ha demonstrates the unsupervised training of a generative RNN to model RL environments through compressed spatial and temporal representations, achieving state of the art results in various environments. His previous works includes Sketch-RNN , a RNN that constructs stroke-based drawings of common objects. It is trained on thousands of human-drawn images representing hundreds of classes, like cats and crabs and the Mona Lisa. Try it out live and have fun!