Solving the Rubik’s cube with deep reinforcement learning and search [pdf]

Note: this is in the new nature machine intelligence journal which many leading academics in the field have argued against for reasons like that:

The article cannot be downloaded: "Complimentary shares do not allow downloading. If you have another access method, please visit the regular publisher article page in order to download."

More background: https://retractionwatch.com/2018/05/01/thousands-boycott-new...

F that. No need to read it either, it's a straight-forward uninteresting "neural networks applied to toy problem X" paper. No interesting use of structure or geometric deep learning, just brute-forcing with 10 billion samples.