This seems like a less sophisticated variant\* of the AlphaGo approach. In both ... | Hacker News

Hacker News new | past | comments | ask | show | jobs | submit

login

shoyer on March 25, 2016 | parent | context | favorite | on: A way to deal with enormous branching factors in s...

This seems like a less sophisticated variant* of the AlphaGo approach. In both cases, Monte Carlo Tree Search (MCTS) is constrained by model for likely future moves. Here, they use a genetic algorithm. In AlphaGo, this is called the "policy network".

* This is not to say that these authors copied AlphaGo -- my uninformed guess is that this general approach of constraining MCTS has been known for a very long time.

jebediah on March 25, 2016 [–]

If I understood correctly, they are not using any MCTS, I believe they are not doing any look ahead, only greedily selecting the best turn.

togelius on March 26, 2016 | [–]

Correct - we are not using MCTS and the approach has nothing to do with AlphaGo.

Join us for AI Startup School this June 16-17 in San Francisco!
Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact