Search found 417 matches

by Rémi Coulom
Wed Dec 06, 2017 6:51 pm
Forum: Computer Chess Club: General Topics
Topic: AlphaZero beats AlphaGo Zero, Stockfish, and Elmo
Replies: 198
Views: 49226

Re: AlphaZero beats AlphaGo Zero, Stockfish, and Elmo

1080 ti is 11.3 TFLOPS: https://www.anandtech.com/show/11172/nvidia-unveils-geforce-gtx-1080-ti-next-week-699 A TPU is 45 TOPS: https://arstechnica.com/information-technology/2017/05/google-brings-45-teraflops-tensor-flow-processors-to-its-compute-cloud/ 1st gen TPU is 92 TOPS and an OP is an 8bit ...
by Rémi Coulom
Wed Dec 06, 2017 6:05 pm
Forum: Computer Chess Club: General Topics
Topic: AlphaZero beats AlphaGo Zero, Stockfish, and Elmo
Replies: 198
Views: 49226

Re: AlphaZero beats AlphaGo Zero, Stockfish, and Elmo

4 hours my ass (pardon my french). Try training it on state-of-the-art 1080. Fully trained network requres 12h on 5000 gen1 TPUs for self-games and 64 gen2 TPUs for training itself. Gen1 TPU is like 30x K80 which is like 5x 1080 in performance. So you'd need like 375k training days with 1080, which...
by Rémi Coulom
Wed Dec 06, 2017 5:43 pm
Forum: Computer Chess Club: General Topics
Topic: AlphaZero beats AlphaGo Zero, Stockfish, and Elmo
Replies: 198
Views: 49226

Re: AlphaZero beats AlphaGo Zero, Stockfish, and Elmo

4 hours my ass (pardon my french). Try training it on state-of-the-art 1080. Fully trained network requres 12h on 5000 gen1 TPUs for self-games and 64 gen2 TPUs for training itself. Gen1 TPU is like 30x K80 which is like 5x 1080 in performance. So you'd need like 375k training days with 1080, which...
by Rémi Coulom
Wed Dec 06, 2017 5:11 pm
Forum: Computer Chess Club: General Topics
Topic: AlphaZero beats AlphaGo Zero, Stockfish, and Elmo
Replies: 198
Views: 49226

Re: AlphaZero beats AlphaGo Zero, Stockfish, and Elmo

4 hours my ass (pardon my french). Try training it on state-of-the-art 1080. Fully trained network requres 12h on 5000 gen1 TPUs for self-games and 64 gen2 TPUs for training itself. Gen1 TPU is like 30x K80 which is like 5x 1080 in performance. So you'd need like 375k training days with 1080, which...
by Rémi Coulom
Wed Dec 06, 2017 2:55 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Google's AlphaGo team has been working on chess
Replies: 203
Views: 63928

Re: Google's AlphaGo team has been working on chess

I have a question that perhaps you can answer right away. Almost a 1000 CPU years went into tuning SF until today.... Would you say that the training of AlphaGo required less or more resources than this? According to the paper, they trained for 9 hours, over 5000 TPUs. 5000 * 9 / 24 = 1875 TPU-days...
by Rémi Coulom
Wed Dec 06, 2017 1:36 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Google's AlphaGo team has been working on chess
Replies: 203
Views: 63928

Re: Google's AlphaGo team has been working on chess

Money would be a better measure. The AlphaZero training system costed $ 4 millions of hardware. (figures given for alpha go zero, don't have source under hand) The paper says they use 5,000 first-generation TPUs, and 64 second-generation TPUs. Such hardware is not available for sale, but might be s...
by Rémi Coulom
Tue Nov 14, 2017 1:40 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Neural networks for chess position evaluation- request
Replies: 23
Views: 6508

Re: Neural networks for chess position evaluation- request

By MCTS, I mean the AlphaGo zero approach, where there is no random playouts. Playouts are completely replaced by neural-network evaluation. I believe a large network can read a lot of tactics. The raw neural-network of AlphaGo is (weak) pro strength by itself, with no search at all. Of course, conv...
by Rémi Coulom
Mon Nov 13, 2017 11:49 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Neural networks for chess position evaluation- request
Replies: 23
Views: 6508

Re: Neural networks for chess position evaluation- request

Another funny idea: it might be possible to make a neural network using bitboards in a binarized neural network: https://arxiv.org/abs/1602.02830
by Rémi Coulom
Mon Nov 13, 2017 11:40 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Neural networks for chess position evaluation- request
Replies: 23
Views: 6508

Re: Neural networks for chess position evaluation- request

I am convinced that a convolutional neural network should be able to learn an evaluation function that is considerably better than any hand-made evaluation. But it will be much slower. Whether it can be competitive with traditional engines is difficult to predict. It is fun and exciting to try, anyw...
by Rémi Coulom
Mon Nov 13, 2017 7:15 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Neural networks for chess position evaluation- request
Replies: 23
Views: 6508

Re: Neural networks for chess position evaluation- request

I took a longer look at what you did. Max pooling is used for image recognition, but makes little sense for chess. You'll need several layers of convolutions before reducing the resolution. I would simply stack a dozen 3x3 convolutions, and then directly go to a couple scalar layers, without any max...