Search found 125 matches

by Rémi Coulom
Sat Jun 20, 2020 9:54 am
Forum: MCTS, RL, and neural networks
Topic: AlphaZero approach for card games
Replies: 1
Views: 100

Re: AlphaZero approach for card games

Hi Alvaro, Because of hidden information, methods that work for board games usually don't work well for card games. The AlphaZero method itself may not work well, but other machine learning algorithms based on self play have been successfully applied to card games. The main approach for statistical ...
by Rémi Coulom
Mon Feb 03, 2020 2:59 pm
Forum: ICGA
Topic: Computers and Games 2020
Replies: 0
Views: 6270

Computers and Games 2020

DESCRIPTION The conference on Computers and Games (CG 2020) will be held in Santiago de Compostela, Spain, in conjunction with the 23rd Computer Olympiad, the World Computer Chess Championship and the ECAI conference. The CG conference is an internationally renowned conference that publishes the st...
by Rémi Coulom
Sun Feb 02, 2020 2:59 pm
Forum: MCTS, RL, and neural networks
Topic: Polygames: Improved Zero Learning
Replies: 0
Views: 6098

Polygames: Improved Zero Learning

github: https://github.com/facebookincubator/polygames paper: https://arxiv.org/abs/2001.09832 abstract: Since DeepMind's AlphaZero, Zero learning quickly became the state-of-the-art method for many board games. It can be improved using a fully convolutional structure (no fully connected layer). Usi...
by Rémi Coulom
Sat Nov 30, 2019 11:41 pm
Forum: Ataxx
Topic: Ataxx in Ludii
Replies: 0
Views: 8110

Ataxx in Ludii

Hi,

Ataxx is included in the latest version of Ludii. You can get it from there:
https://ludii.games/download.php

Menu: File/Load Game -> Games/board/space/territory/Ataxx

The GUI lets you play against an UCT player. I played a game, and won easily.

The alphabeta player is much stronger.

Rémi
by Rémi Coulom
Wed Nov 20, 2019 10:54 pm
Forum: MCTS, RL, and neural networks
Topic: TD(1)
Replies: 0
Views: 6487

TD(1)

I'll try to explain TD(1) as simply as possible. The idea is to tune the evaluation function so that it predicts game outcomes as accurately as possible. Depending on the game, the nature of the game outcome might be a bit different. Some Othello programs try to predict and maximize the final score....
by Rémi Coulom
Sat Nov 09, 2019 10:19 pm
Forum: MCTS, RL, and neural networks
Topic: My home-made CUDA kernel for convolutions
Replies: 0
Views: 7694

My home-made CUDA kernel for convolutions

Hi, I wrote a CUDA kernel for convolution, and it outperforms cuDNN for small tensors. cuDNN is efficient for large batches, but slow for small batches. Here are some numbers for my Go program. This is a network with 80 layers of 128 channels, 19x19 board, V100 (PCIExpress), half-precision, NHWC ten...
by Rémi Coulom
Mon Nov 04, 2019 6:50 pm
Forum: Ataxx
Topic: Crazy Zero
Replies: 0
Views: 6816

Crazy Zero

Hi, Here is a short report about my Crazy Zero experiments. Crazy Zero is a generic implementation of AlphaZero. I already used it for Go, shogi, gomoku, renju, othello, and chess. The version that played on the Ataxx server (http://server.ataxx.org/) is less than one day old. The network architectu...
by Rémi Coulom
Thu Oct 31, 2019 12:19 pm
Forum: General Discussion
Topic: AlphaStar: Grandmaster level in StarCraft II using multi-agent reinforcement learning
Replies: 0
Views: 10860

AlphaStar: Grandmaster level in StarCraft II using multi-agent reinforcement learning

https://www.deepmind.com/blog/article/AlphaStar-Grandmaster-level-in-StarCraft-II-using-multi-agent-reinforcement-learning Direct link to the paper: https://storage.googleapis.com/deepmind-media/research/alphastar/AlphaStar_unformatted.pdf The most interesting contribution seems to be their league s...
by Rémi Coulom
Sat Oct 26, 2019 10:38 am
Forum: Game-specific references
Topic: Mahjong
Replies: 1
Views: 16533

Re: Mahjong

Microsoft has produced a very strong AI:
https://medium.com/syncedreview/meet-mi ... b0a63eb871
Very little technical details, unfortunately.
by Rémi Coulom
Wed Oct 23, 2019 5:22 pm
Forum: ICGA
Topic: A Welcome from the new President, Jonathan Schaeffer
Replies: 0
Views: 7277

A Welcome from the new President, Jonathan Schaeffer

https://icga.org/?page_id=2984

Thanks Jonathan for accepting to become the new President of the ICGA!