It is currently Thu Mar 28, 2024 7:31 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 101 posts ]  Go to page 1, 2, 3, 4, 5, 6  Next
Author Message
Offline
 Post subject: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-0
Post #1 Posted: Wed Jan 27, 2016 10:55 am 
Lives in gote

Posts: 653
Location: Austin, Texas, USA
Liked others: 54
Was liked: 216
Incredible news, I really don't quite believe it... But it looks pretty legit so far.

http://wayt.synology.me/wordpress/1348-2
https://www.youtube.com/watch?v=g-dKXOlsf98
http://deepmind.com/alpha-go.html
https://storage.googleapis.com/deepmind ... ing-go.pdf

AlphaGo played a 5 game match with Fan Hui and won 5-0. 19x19 board, no handicaps(!)

From the video it sounds like their major innovation is to use two neural nets instead of just one. One is a 'policy network' to select move candidates, this is what Facebook did. But then they created a 'value network' that evaluates leaf nodes. Other bots had been using Monte Carlo playouts for their evaluation.

- Kim Myungwan will be doing a review of the AlphaGo vs Fan Hui 2p games on the AGA Youtube this Friday.
- Google DeepMind, the British artificial intelligence company which developed AlphaGo, has issued a $1 million challenge to Lee Sedol from South Korea, the top player in the world for much of the last 10 years, to play in March.


Last edited by yoyoma on Wed Jan 27, 2016 1:49 pm, edited 4 times in total.

This post by yoyoma was liked by 3 people: emeraldemon, hyperpape, xiangz
Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #2 Posted: Wed Jan 27, 2016 11:05 am 
Dies with sente
User avatar

Posts: 101
Location: Geneva, Switzerland
Liked others: 66
Was liked: 57
Rank: DGS 6k
KGS: Codexus
DGS: Codexus
This is huge. I hope somebody with Nature access can give us some details about under what conditions the games were played.

EDIT: Never mind the paper is available for free from Google's website:
http://deepmind.com/alpha-go.html
https://storage.googleapis.com/deepmind ... ing-go.pdf


Last edited by Codexus on Wed Jan 27, 2016 11:08 am, edited 1 time in total.

This post by Codexus was liked by 2 people: emeraldemon, Nikolas73
Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #3 Posted: Wed Jan 27, 2016 11:06 am 
Lives in gote

Posts: 436
Liked others: 1
Was liked: 38
Rank: KGS 5 kyu
yoyoma wrote:
Incredible news, I really don't quite believe it... But it looks pretty legit so far.

http://wayt.synology.me/wordpress/1348-2
https://www.youtube.com/watch?v=g-dKXOlsf98
http://www.nature.com/nature/journal/v5 ... 16961.html (actual paper is behind a paywall)

AlphaGo played a 5 game match with Fan Hui and won 5-0. 19x19 board, no handicaps(!)

From the video it sounds like their major innovation is to use two neural nets instead of just one. One is a 'policy network' to select move candidates, this is what Facebook did. But then they created a 'value network' that evaluates leaf nodes. Other bots had been using Monte Carlo playouts for their evaluation.


:O

What????

Ok if it defeats Lee Sedol then :salute:

Google is my hero!

https://www.youtube.com/watch?v=SUbqykXVx0A

Another video.


Last edited by Krama on Wed Jan 27, 2016 11:07 am, edited 1 time in total.
Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #4 Posted: Wed Jan 27, 2016 11:24 am 
Lives in sente

Posts: 827
Location: UK
Liked others: 568
Was liked: 84
Rank: OGS 9kyu
Universal go server handle: WindnWater, Elom
It's an interesting experience seeing videos of Fan Hui 2p and Matuesz Surma 1p playing go on national television. I think that comes rather close to a satisfying verification.

_________________
On Go proverbs:
"A fine Gotation is a diamond in the hand of a dan of wit and a pebble in the hand of a kyu" —Joseph Raux misquoted.

Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #5 Posted: Wed Jan 27, 2016 11:28 am 
Dies with sente
User avatar

Posts: 101
Location: Geneva, Switzerland
Liked others: 66
Was liked: 57
Rank: DGS 6k
KGS: Codexus
DGS: Codexus
From the paper, the game time was 1 hour main time + 30s byo-yomi. This is no blitz trick.

Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #6 Posted: Wed Jan 27, 2016 11:40 am 
Oza

Posts: 2180
Location: ʍoquıɐɹ ǝɥʇ ɹǝʌo 'ǝɹǝɥʍǝɯos
Liked others: 237
Was liked: 662
Rank: AGA 5d
GD Posts: 4312
Online playing schedule: Every tenth February 29th from 20:00-20:01 (if time permits)
I'd be curious to know how much they had to pay Lee to participate in this.

_________________
Still officially AGA 5d but I play so irregularly these days that I am probably only 3d or 4d over the board (but hopefully still 5d in terms of knowledge, theory and the ability to contribute).

Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #7 Posted: Wed Jan 27, 2016 11:42 am 
Lives with ko

Posts: 199
Liked others: 6
Was liked: 55
Rank: KGS 3 kyu
I read $1 million.

Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #8 Posted: Wed Jan 27, 2016 11:55 am 
Tengen

Posts: 4380
Location: North Carolina
Liked others: 499
Was liked: 733
Rank: AGA 3k
GD Posts: 65
OGS: Hyperpape 4k
One stat from someone who worked on the project (https://news.ycombinator.com/item?id=10982008): they can give Crazy Stone a 4 stone handicap and win 80% of the time. I'd still suppose they're an underdog against Lee Sedol, but maybe not a hopeless one?

_________________
Occupy Babel!

Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #9 Posted: Wed Jan 27, 2016 12:01 pm 
Lives in gote

Posts: 448
Liked others: 5
Was liked: 187
Rank: BGA 3 dan
Marcel Grünauer wrote:
I really hope this technology will trickle down to the desktop or to mobile devices.


I asked Demis Hassabis about processor power, and it might be 200 cores. So, a little while yet.

Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #10 Posted: Wed Jan 27, 2016 12:03 pm 
Lives in sente

Posts: 827
Location: UK
Liked others: 568
Was liked: 84
Rank: OGS 9kyu
Universal go server handle: WindnWater, Elom
Marcel Grünauer wrote:
Elom wrote:
It's an interesting experience seeing videos of Fan Hui 2p and Matuesz Surma 1p playing go on national television. I think that comes rather close to a satisfying verification.


Where did you see Mateusz Surma playing on national TV?


http://www.bbc.co.uk/news/technology-35420579

Ranked 3rd on top stories, but those ranks change all the time I guess (a lot like my ranks).

_________________
On Go proverbs:
"A fine Gotation is a diamond in the hand of a dan of wit and a pebble in the hand of a kyu" —Joseph Raux misquoted.


Last edited by Elom on Wed Jan 27, 2016 12:12 pm, edited 1 time in total.
Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #11 Posted: Wed Jan 27, 2016 12:08 pm 
Gosei
User avatar

Posts: 1744
Liked others: 702
Was liked: 288
KGS: greendemon
Tygem: greendemon
DGS: smaragdaemon
OGS: emeraldemon
Super exciting news, both because of the AI development and also because it may raise some hype and awareness about Go in the US.

I couldn't resist looking up Fan Hui on Remi's goratings: http://www.goratings.org/players/1480.html

He's at 2916 elo, and Lee Sedol is at 3515, so by rating Fan Hui would have about a 3% chance of beating Lee Sedol. This seems about in line with the team jubango Lee played last year, where he beat top westerners at 2 stones. So beating Fan Hui is definitely an accomplishment, but Lee Sedol represents a significant step up in difficulty for AlphaGo.

That said, I doubt anyone predicted we would see a match like this so soon, Go AI is developing incredibly quickly. Is it too much to ask that google would make the AI open source? I'll have to read the paper...

Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #12 Posted: Wed Jan 27, 2016 12:14 pm 
Lives in gote

Posts: 436
Liked others: 1
Was liked: 38
Rank: KGS 5 kyu
The games.




Attachments:
AlphaGo_FanHui_game_3.sgf [1.14 KiB]
Downloaded 2198 times
AlphaGo_FanHui_game_2.sgf [1.22 KiB]
Downloaded 2264 times
AlphaGo_FanHui_game_1.sgf [1.6 KiB]
Downloaded 2272 times

This post by Krama was liked by 3 people: Bonobo, Charles Matthews, emeraldemon
Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #13 Posted: Wed Jan 27, 2016 12:16 pm 
Lives in sente

Posts: 902
Location: Fort Collins, CO
Liked others: 319
Was liked: 287
Rank: AGA 3k
Universal go server handle: jeromie
Codexus wrote:
From the paper, the game time was 1 hour main time + 30s byo-yomi. This is no blitz trick.


According to the paper, Fan Hui won two out of 5 informal games played at shorter time controls. If anything, it seems the computer plays better at standard time controls.

Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #14 Posted: Wed Jan 27, 2016 12:16 pm 
Lives in gote

Posts: 436
Liked others: 1
Was liked: 38
Rank: KGS 5 kyu
4th and 5th




Attachments:
AlphaGo_FanHui_game_4.sgf [1.12 KiB]
Downloaded 2254 times
AlphaGo_FanHui_game_5.sgf [1.4 KiB]
Downloaded 2260 times

This post by Krama was liked by: Bonobo
Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #15 Posted: Wed Jan 27, 2016 12:21 pm 
Lives in sente

Posts: 902
Location: Fort Collins, CO
Liked others: 319
Was liked: 287
Rank: AGA 3k
Universal go server handle: jeromie
Charles Matthews wrote:
Marcel Grünauer wrote:
I really hope this technology will trickle down to the desktop or to mobile devices.


I asked Demis Hassabis about processor power, and it might be 200 cores. So, a little while yet.


According to the paper, the distributed version of AlphaGo was running on 1202 CPUs and 176 GPUs. Their non-distributed version was running on a machine with 48 CPUs and 8 GPUs. So yeah, this isn't off the shelf hardware. But if CPU power is the only limiting factor, I think we can expect some VERY strong programs available for home computers in the near future.

Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #16 Posted: Wed Jan 27, 2016 1:28 pm 
Lives in sente

Posts: 923
Location: UK
Liked others: 72
Was liked: 479
Rank: 5 dan
KGS: macelee
I just briefly went through all five games. My observations
- Fan Hui's openings are very traditional and peaceful. AI probably knows all these openings very well from past pro games. So there is nothing human player can gain at this stage. It remains to be seem how the AI reacts to the more 'modern' and 'imaginative' ways younger Chinese and Korean players play (for example how would the AI reacts to situations as in this game: http://www.go4go.net/go/games/sgfview/52548 (moves 28 through 68).
- Fan appears to make a number of mid-game mistakes. But so does the AI player (for example, one obvious mistake was move 60 in game 3).

Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #17 Posted: Wed Jan 27, 2016 1:37 pm 
Gosei
User avatar

Posts: 1744
Liked others: 702
Was liked: 288
KGS: greendemon
Tygem: greendemon
DGS: smaragdaemon
OGS: emeraldemon
macelee wrote:
I just briefly went through all five games. My observations
- Fan Hui's openings are very traditional and peaceful. AI probably knows all these openings very well from past pro games. So there is nothing human player can gain at this stage. It remains to be seem how the AI reacts to the more 'modern' ways younger Chinese and Korean players play (for example how would the AI reacts to situations as in this game: http://www.go4go.net/go/games/sgfview/52548 (moves 28 through 68).
- Fan appears to make a number of mid-game mistakes. But so does the AI player (for example, one obvious mistake was move 60 in game 3).


It definitely seems to pull openings from pro games, which makes sense. It played the large avalanche perfectly, and the joseki after microchinese - split - attach that Fan Hui used both times. I also noticed AlphaGo played nirensei both games as black, making a sanrensei for one of them. It's interesting to me because black nirensei isn't that popular with pros today.

AlphaGo does vary its openings at least some, since it faced an identical position and chose the avalanche for one game and the hane for the other.

Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #18 Posted: Wed Jan 27, 2016 1:38 pm 
Lives in gote

Posts: 436
Liked others: 1
Was liked: 38
Rank: KGS 5 kyu
Would be interesting to see strong pros review this game.

This is probably a high insei level game in asia.

Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #19 Posted: Wed Jan 27, 2016 1:46 pm 
Lives in gote

Posts: 653
Location: Austin, Texas, USA
Liked others: 54
Was liked: 216
Some quote from the paper
https://storage.googleapis.com/deepmind ... ing-go.pdf

Quote:
We trained the policy network pσ to classify positions according to expert moves played in the KGS data set. This data set contains 29.4 million positions from 160,000 games played by KGS 6 to 9 dan human players; 35.4% of the games are handicap games.


Quote:
The final version of AlphaGo used 40 search threads, 48 CPUs, and 8 GPUs. We also implemented a distributed version of AlphaGo that exploited multiple machines, 40 search threads, 1202 CPUs and 176 GPUs.


Quote:
The results of the tournament (see Figure 4,a) suggest that single machine AlphaGo is many dan ranks stronger than any previous Go program, winning 494 out of 495 games (99.8%) against other Go programs. To provide a greater challenge to AlphaGo, we also played games with 4 handicap stones (i.e. free moves for the opponent); AlphaGo won 77%, 86%, and 99% of handicap games against Crazy Stone, Zen and Pachi respectively. The distributed version of AlphaGo was significantly stronger, winning 77% of games against single machine AlphaGo and 100% of its games against other programs.


AlphaGo was using 48 CPUs + 8 GPUs. CrazyStone had 32 CPUs, Zen had 8 CPUs, Pachi had 16 CPUs. They were using 5 seconds per move.

Top
 Profile  
 
Offline
 Post subject: Re: Google's AlphaGo defeats Fan Hui 2p, 19x19, no handi, 5-
Post #20 Posted: Wed Jan 27, 2016 2:25 pm 
Lives in gote

Posts: 448
Liked others: 5
Was liked: 187
Rank: BGA 3 dan
macelee wrote:
Fan appears to make a number of mid-game mistakes. But so does the AI player (for example, one obvious mistake was move 60 in game 3).


I wondered about that. Playing out ko threats is usually poor, especially when they lose endgame points. But "making the next threat bigger" can be justified. (There was a ko top right, and I don't suppose it was over the analysis horizon.)

But I doubt we're going to be able to get inside the head of this type of algorithm, with our usual concepts. Over the five games it appears to be at home in the middlegame.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 101 posts ]  Go to page 1, 2, 3, 4, 5, 6  Next

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group