In this paper insight we discuss two recent papers that deal with the following problem: how to animate an image given a target video with motion in a self-supervised manner. This involves unsupervised keypoint detection, segmentation and optical flow estimation along the way.

We trained a Deep RL agent for our 1 v 1 real-time action strategy game, Heroic - Magic Duel, via ensemble self-play with only a simple reward of +-1. The agent obtains over 50% win rate against a top human player and over 60% against the existing AI.

In this post, we introduce some theory of contextual bandits, which have become a popular approach to recommender problems, where there are many possible items to offer and there is frequent interaction with the users, e.g. via click/no-click or purchase/no-purchase events.

In this post, we give an overview of some of the recent advancements in deep learning-based character animation. In particular, we discuss two papers that provide a way to use neural networks to turn motion capture data into interactive character controllers.