<html><head></head><body><div>Google DeepMind's paper on reinforcement learning playing Atari games is now out in Nature: Mnih, V., Kavukcuoglu, K., Silver, D., Graves, A., Antonoglou, I., Wierstra, D., & Riedmiller, M. (2013). Playing atari with deep reinforcement learning. arXiv preprint arXiv:1312.5602.</div><div><br></div>http://googleresearch.blogspot.co.uk/2015/02/from-pixels-to-actions-human-level.html<div>http://www.nature.com/nature/journal/v518/n7540/full/nature14236.html</div><div>https://www.youtube.com/watch?v=iqXKQf2BOSE</div><div>http://arxiv.org/abs/1312.5602</div><div><br></div><div>The nature version has a really interesting plot of relative performance on different games. It looks like the system is amazing at games where the current state is all you need to deal with, while it is less successful at games where you need to find objects and use them in the right location at the state of the game. Not too surprising (reinforcement learning is closely linked to Markov chains) but nevertheless a good indication of where the next rewards in research are likely to lie. </div><div><br></div><div><br><br>Anders Sandberg,
Future of Humanity Institute
Philosophy Faculty of Oxford University</div></body></html>