Advertisement
Advertisement

More Related Content

Advertisement

Programs that Play better than Us

  1. Programs that Play better than Us Melvin Zhang melvin@melvinzhang.net @melvinzhangzy
  2. https://en.wikipedia.org/wiki/File:ST Battle Chess.png
  3. https://en.wikipedia.org/wiki/Deep Blue (chess computer) Deep Blue (IBM, 1996)
  4. http://afflictor.com/2012/09/11/chess-programs-regularly-play-at-good-amateur-level/
  5. Game tree
  6. Optimal play Terminal min player max player
  7. Optimal play 1 01 1 1Terminal min player max player
  8. Optimal play 1 01 1 1 0 Terminal min player max player
  9. Optimal play 1 01 1 1 0 1 Terminal min player max player
  10. Optimal play 1 01 1 1 0 1 1 Terminal min player max player
  11. Chess has about 1046 states!
  12. Minimax algorithm Cut-off min player max player
  13. Minimax algorithm .7 .1 .6 .9Cut-off min player max player
  14. Minimax algorithm .7 .1 .6 .9 .1 Cut-off min player max player
  15. Minimax algorithm .7 .1 .6 .9 .1 .6 Cut-off min player max player
  16. Minimax algorithm .7 .1 .6 .9 .1 .6 .6 Cut-off min player max player
  17. https://stockfishchess.org/ Stockfish
  18. https://tests.stockfishchess.org/ Testing AI changes is crucial
  19. Value functions are hard!
  20. http://mathworld.wolfram.com/Go.html
  21. http://www.remi-coulom.fr/CrazyStone/ R´emi Coulom
  22. http://www.wired.com/2014/05/the-world-of-computer-go/
  23. Monte Carlo evaluations Cut-off min player max player
  24. Monte Carlo evaluations Cut-off min player max player
  25. Monte Carlo evaluations Cut-off min player max player
  26. Monte Carlo evaluations Cut-off min player max player
  27. Monte Carlo evaluations Cut-off min player max player .7
  28. Monte Carlo Tree Search (MCTS)
  29. by Google Deepmind https://deepmind.com/research/alphago/
  30. https://gogameguru.com/alphago-races-ahead-2-0-lee-sedol/
  31. MCTS + Policy and value networks
  32. http://www.nature.com/nature/journal/v529/n7587/full/nature16961.html MCTS + Policy and value networks
  33. Some games have hidden information!
  34. http://magic.wizards.com/en/events/coverage/gpsin15/father-son-2015-06-27
  35. https://magarena.github.io
  36. Determinization: choose a random instance of the hidden information during simulation
  37. Comparison of Minimax and MCTS At 1s thinking time: Minimax MCTS 1 0.88
  38. Comparison of Minimax and MCTS At 1s thinking time: Minimax MCTS 1 0.88 At 4s thinking time: Minimax MCTS 1 1.71
  39. Open problems
  40. MCTS is bad at tight tactical play.
  41. MCTS plays badly when it is behind in the game.
  42. Further readings
  43. Further readings
Advertisement