"Ingo Althöfer" | 6 Feb 13:40 2016
Picon
Picon

Match Date: March 09 - 15

Hello,

on the Alpha-Go website a date for the match between
Lee Sedol and Alpha-Go is given:
5 rounds, to be played in Seoul on

* Wednesday, March 09
* Thursday, March 10

* Saturday, March 12
* Sunday, March 13

* Tuesday, March 15

It seems that March 11 and 14 are rest days.

http://deepmind.com/alpha-go.html

Ingo.
_______________________________________________
Computer-go mailing list
Computer-go <at> computer-go.org
http://computer-go.org/mailman/listinfo/computer-go
Marc Landgraf | 4 Feb 12:52 2016
Picon

Neural Net move prediction

Hi,

lately a friend and me wondered about the following idea.

Let's assume you have a reasonably strong move prediction DCNN. What
happens if you now train a second net on the same database.
When training the first net, you tried to maximize the judgement value
of the expert move. But for the second net you now try to maximize the
maximum of the judgement of both nets. This means, that the second net
does not profit from finding moves the first net can easily find, but
instead will try to fill in the weaknesses of the first net.
In practical application the easy static usage would be to first
expand the top2 candidates of the first net, then mix in the top
candidate of the second net, then again the next 2 candidates from the
first net, etc.

What do you guys think about that?

Cheers, Marc
_______________________________________________
Computer-go mailing list
Computer-go <at> computer-go.org
http://computer-go.org/mailman/listinfo/computer-go
Andrés Román | 4 Feb 00:09 2016
Picon
Gravatar

Forecasting Lee Sedol vs. AlphaGo

Hello !

I would like to share with this list the possibility to participate forecasting on the forecoming game,

"Will Google's AlphaGo beat world champion Lee Sedol in the five game Go match planned for March 2016?"

That is posted in the Good Judgement Project open forecasting site.
It will be great to read the forecasts and opinions of members of this list.

Andres
_______________________________________________
Computer-go mailing list
Computer-go <at> computer-go.org
http://computer-go.org/mailman/listinfo/computer-go
Robert Jasiek | 3 Feb 10:51 2016
Picon

Knowledge Details

The current fashion favours general AI approaches forgoing knowledge 
details. Given enough calculation power applied to well chosen AI 
techiques, many knowledge details are redundant because they are 
generated automatically: AlphaGo does play (at least some) ko fights 
with ko threats, tesujis, test moves, (at least some) life and death or 
semeai problems etc. At the same time, AI calculation power is still not 
large enough to generate all human knowledge details. Aji with long-term 
impact and maintaining the life status "independently alive" instead of 
unnecessarily transforming it to "(ko|independently alive)" (aka 
"unsettled") are prime examples. Programs also play for the win 
regardless of whether moves are suboptimal for the score difference - 
human players tend to avoid such (programs would also profit from 
avoiding such to prevent losing when making a later mistake due to a 
knowledge gap related to insufficient error handling). There is another 
great threat related to knowledge details, which is not immediately 
apparent and will be even much less apparent when programs will exceed 
top human playing strength: A program can run into a situation where an 
infrequent knowledge detail becomes relevant. And a program can run into 
ordinary software or hardware bugs, something that must be detected and 
correct on the AI level.

My conclusion is: human expert knowledge on details of go theory matters.

There have been 9p players committing self-atari when filling a dame, so 
you might argue that programs may infrequently make similar blunders. 
When I issued a million dollar prize, I'd prefer human expert knowledge 
implemented at least as an additional layer of error handling.

(Other fun includes internet connection trouble, server bugs of 
distributed computers, hardware bugs of the local interface computers or 
interrupted power supply.)

--

-- 
robert jasiek
_______________________________________________
Computer-go mailing list
Computer-go <at> computer-go.org
http://computer-go.org/mailman/listinfo/computer-go
Robert Jasiek | 3 Feb 10:24 2016
Picon

57%

AlphaGo is said to predict 57% of professionals' moves. How is this 
number measured and from which sample?

At some turns, there is only one correct move - at other turns, strong 
go players would say that there are several valid supposedly correct 
moves. This is one of the reasons why 100% cannot be the optimum but a 
smaller percentage must be the best.

Pro players, or players of the database sample (incl. real world 3d 
players being 9d on KGS), make mistakes. A neural net learns from a 
sample and therefore also learns the mistakes. This is the most 
important reason why 100% cannot be the optimum but a smaller percentage 
must be the best.

(Roughly) which percentage is optimal? Why? Is the optimum greater or 
smaller than 57%?

--

-- 
robert jasiek
_______________________________________________
Computer-go mailing list
Computer-go <at> computer-go.org
http://computer-go.org/mailman/listinfo/computer-go
Hiroshi Yamashita | 2 Feb 22:33 2016
Picon

guess AlphaGo's CGOS rating

Hi,

I made CGOS rating include AlphaGo.

                 CGOS    Remi's pro rating(same as paper)

Ke Jie           4642?   3620  Strongest human
Lee Sedol        4538?   3516
AlphaGo          4162?   3140  Distributed, 1202CPUs, 176GPUs
Fan Hui          3942?   2920
AlphaGo          3912?   2890  48CPUs, 8GPUs, one machine
Zen 24core       3800???       KGS 7d, 24 cores, one machine.
Zen-10.8-2c      3485          
CrazyStone-0002  3313          i7-5930K, 6 threads
Zen-10.8-1c      3299          DCNN versoin
Zen-1c-2.8G      3072
AlphaGo RL       2780?         Reinforcement learning, no search
Aya786l_10k      2718          KGS 2d
darkfores2       2657?         KGS 3d, no search
AlphaGo SL       2539?   1517  DCNN,   no search
darkfores1       2490?         KGS 2d, no search
pachi11_Pat_100k 2478
darkforest       2271?         KGS 1d, no search
DCNN_Detlef54    2179          Detlef's 54%, no search
pachi11_Pat_10k  2104
DCNN-Detlef      1903          Detlef's 44%, no search
Gnugo-3.7.10-a1  1800          KGS 5k

? is guess. ??? is my anticipation.

This result is based on AlphaGo(RL) beats Pachi 100k with 85% winrate.
darkforest is also based on winrate against Pachi 100k.
Pro and computers comparison is maybe only from 8-2 result AlphaGo vs
 Fan Hui. So pro is maybe more strong (or weak).
Watching this table, Zen 7d maybe play nice game against one machine
 AlphaGo.

Pro rating is from Remi's site.
http://www.goratings.org/
CGOS rating is from BayesElo.
http://www.yss-aya.com/cgos/19x19/bayes.html

Regards,
Hiroshi Yamashita

_______________________________________________
Computer-go mailing list
Computer-go <at> computer-go.org
http://computer-go.org/mailman/listinfo/computer-go
David Fotland | 2 Feb 19:25 2016

What hardware to use to train the DNN

Detlef, Hiroshi, Hideki, and others,

I have caffelib integrated with Many Faces so I can evaluate a DNN.  Thank you very much Detlef for sample code
to set up the input layer.  Building caffe on windows is painful.  If anyone else is doing it and gets stuck I
might be able to help.

What hardware are you using to train networks?  I don’t have a cuda-capable GPU yet, so I'm going to buy a new
box.  I'd like some advice.  Caffe is not well supported on Windows, so I plan to use a Linux box for training,
but continue to use Windows for testing and development.  For competitions I could use either windows or linux.

Thanks in advance,

David

> -----Original Message-----
> From: Computer-go [mailto:computer-go-bounces <at> computer-go.org] On Behalf
> Of Hiroshi Yamashita
> Sent: Monday, February 01, 2016 11:26 PM
> To: computer-go <at> computer-go.org
> Subject: *****SPAM***** Re: [Computer-go] DCNN can solve semeai?
> 
> Hi Detlef,
> 
> My study heavily depends on your information. Especially Oakfoam code,
> lenet.prototxt and generate_sample_data_leveldb.py was helpful. Thanks!
> 
> > Quite interesting that you do not reach the prediction rate 57% from
> > the facebook paper by far too! I have the same experience with the
> 
> I'm trying 12 layers 256 filters, but it is around 49.8%.
> I think 57% is maybe from KGS games.
> 
> > Did you strip the games before 1800AD, as mentioned in the FB paper? I
> > did not do it and was thinking my training is not ok, but as you have
> > the same result probably this is the only difference?!
> 
> I also did not use before 1800AD. And don't use hadicap games.
> Training positions are 15693570 from 76000 games.
> Test     positions are   445693 from  2156 games.
> All games are shuffled in advance. Each position is randomly rotated.
> And memorizing 24000 positions, then shuffle and store to LebelDB.
> At first I did not shuffle games. Then accuracy is down each 61000
> iteration (one epoch, 256 mini-batch).
> http://www.yss-aya.com/20160108.png
> It means DCNN understands easily the difference 1800AD games and  2015AD
> games. I was surprised DCNN's ability. And maybe 1800AD games  are also
> not good for training?
> 
> Regards,
> Hiroshi Yamashita
> 
> ----- Original Message -----
> From: "Detlef Schmicker" <ds2 <at> physik.de>
> To: <computer-go <at> computer-go.org>
> Sent: Tuesday, February 02, 2016 3:15 PM
> Subject: Re: [Computer-go] DCNN can solve semeai?
> 
> > Thanks a lot for sharing this.
> >
> > Quite interesting that you do not reach the prediction rate 57% from
> > the facebook paper by far too! I have the same experience with the
> > GoGoD database. My numbers are nearly the same as yours 49% :) my net
> > is quite simelar, but I use 7,5,5,3,3,.... with 12 layers in total.
> >
> > Did you strip the games before 1800AD, as mentioned in the FB paper? I
> > did not do it and was thinking my training is not ok, but as you have
> > the same result probably this is the only difference?!
> >
> > Best regards,
> >
> > Detlef
> 
> _______________________________________________
> Computer-go mailing list
> Computer-go <at> computer-go.org
> http://computer-go.org/mailman/listinfo/computer-go

_______________________________________________
Computer-go mailing list
Computer-go <at> computer-go.org
http://computer-go.org/mailman/listinfo/computer-go
Mari Glambert | 2 Feb 05:21 2016
Picon

Last CFP: GCIRE2016 | GREEN COMPUTING, INTELLIGENT AND RENEWABLE ENERGIES | Philippines

THE SECOND INTERNATIONAL CONFERENCE ON GREEN COMPUTING, INTELLIGENT AND RENEWABLE ENERGIES (GCIRE2016)

Where: University of Perpetual Help System DALTA, Las Piñas, Manila, Philippines
When: February 24-26, 2016

Publication: The Society of Digital Information and Wireless Communications Digital Library, conference proceedings, other indexing sites and special issues on (1) International Journal of New Computer Architectures and their Applications; (2) International Journal of Digital Information and Wireless Communications; International Journal of Cyber-Security and Digital Forensics; (4) International Journal of E-Learning and Educational Technologies in the Digital Media.

Keynote Speaker: Dr. Ekaterina Pshehotskaya, InfoWatch, Russia
Keynote Title: Linguistics in Data Loss Prevention 

Full Paper Submission Deadline: February 5, 2016

Accepted Paper/s Guidelines Link: http://bit.do/gcire2016-guidelines
Files Required: Final Paper and Copyright

First Paper Registration: $390
Second Paper (or more) Registration: $290
Guest Registration: $250

Paper Registration Deadline: February 14, 2016
Guest Registration Deadline: February 24, 2016

General Chair: Engr. Lorena C. Ilagan

Co-event Conferences:
(1) The International Conference on Innovations in Intelligent Systems and Computing Technologies (ICIISCT2016)
(2) The Second International Conference on Electrical and Electronic Engineering, Telecommunication Engineering, and Mechatronics (EEETEM2016)

KINDLY FORWARD THIS ANNOUNCEMENT TO COLLEAGUES WHO MIGHT BE OF INTEREST. GOOD DAY!

_______________________________________________
Computer-go mailing list
Computer-go <at> computer-go.org
http://computer-go.org/mailman/listinfo/computer-go
Igor Polyakov | 2 Feb 04:07 2016
Picon

Re: AlphaGo and the Standard Mistake in Research and Journalism

Have you heard of https://en.wikipedia.org/wiki/Graham's_number ? It is 
certainly far too large to write out the digits to, in fact the number 
of digits in that number is too large to write

On 2016-01-31 8:19, John Tromp wrote:
> dear Robert,
>
>> The number G19 of legal games under a given go ruleset is unknown.
> It will never be known since there's not enough space in the known
> universe to write it down. We're talking about a number with over
> 10^100 digits.
>
>> For positional
>> superko (prohibition of recreation of the same position after
>> completion of a move on the board), no passes, and no resignation, the
>> number of possible games is smaller than N^P19
> The no-pass restriction makes this rather uninteresting.
> But actually, the same bound applies to games with passes,
> stated as Theorem 7 in our paper.
>
> Besides this upper bound, I can think of only two other numbers
> that are well defined and interesting, namely
>
> 1) the size of the smallest search tree that proves the perfect komi.
> and
> 2) same but for a full-width tree
>
> These are called the decision complexity and game tree complexity on
> https://en.wikipedia.org/wiki/Game_complexity#Decision_complexity
>
> It is reasonable to expect the perfect komi does not depend on games
> of more than 361 moves. Even with some ko fights, the ko recaptures
> are likely bounded by the number of unplayed points.
> The branching factor will be high though; let's put it at at most 200
> on (geometric) average.
>
> Then we estimate the decision complexity to be upper bounded by
> 200^181 and the game tree complexity by 200^361.
>
> regards,
> -John
> _______________________________________________
> Computer-go mailing list
> Computer-go <at> computer-go.org
> http://computer-go.org/mailman/listinfo/computer-go

_______________________________________________
Computer-go mailing list
Computer-go <at> computer-go.org
http://computer-go.org/mailman/listinfo/computer-go
Hiroshi Yamashita | 2 Feb 03:25 2016
Picon

DCNN can solve semeai?

Hi,

I made DCNN, and tried whether DCNN can understand semeai.

1. try one playout that always select DCNN highest probablity move.
2. try 100 playouts that select moves from DCNN probability.
   (one playout takes 4 seconds.)

Result is DCNN does not understand semeai. It can play semeai like moves,
 but far from perfect. Maybe need more difficult feature?

           DCNN highest    DCNN 100 playouts     Aya playout
problem1       0                 53                   20
problem2       0                 54                   85
problem3     100                 91                   70
problem4     100                 66                   66
problem5       0                 51                   62
problem6       0                 45                   67
problem7     100                 95                   90
problem8       0                  9                   90
problem9       0                 39                   50
--------------------------------------------------------
Average       33                 56                   67

100 is correct. 0 is wrong.
 53 means DCNN playout is correct 53 times out of 100 playouts.

DCNN is 
  12 layers, 128 filters. (5x5_128, 3x3_128 x10, 3x3_1)
  It predicts next 3 moves, like Facebook paper.
  Test accuracy is next_1 49%, next_2 27%, next_3 16%.
  http://www.yss-aya.com/20160123_3steps.png
  Feature are
    liberty black and white. 1,2,3,4>=
    stones black and white. 1,2,3,4>=
    previous move 1,2,3,4,5
    previous ko 1,2,3
    CFG Distance 1,2,3,4,5>=
    string life and death by search. dead, killed next, kill move,
     life move. 8 planes
    group life and death. It is from Aya's classic evaluation(KGS 8k).
     6 planes
    territory. black and white.
    half eye. black and white.
    recapture soon. (if black play here, recapture soon). black and white
  All 49 channels.
  Learning games are from GoGoD, 78000 games.
  This DCNN runs on CGOS by DCNN_Aya_i49_a49, no search. GTX 980.
  Its winrate 40% against Pachi 100k.
  AlphaGo DCNN(RL) winrate is 85% against Pachi 100k.
  So this DCNN is weaker than AlphaGo(RL) by 370 Elo,
              and weaker than AlphaGo(SL) by 129 Elo.

Problem8, W lives. 4 libs vs 5 libs (top left)
Problem9, W lives. 5 libs vs 6 libs (bottom left)

19.O.X.XO.OX.........   DCNN answered badly these two problem.
18OXXXXXO.XX.........
17.OOOOOO.X.....O.O..   Left Top is 4 libs vs 5 libs
16OO.XXXXX.......O...   White(O) must live
15.XX................   
14X...............O..   DCNN answer is White(O) lives 9%
13...................   DCNN best move playout also fails.
12...................
11................O..
10...................
 9................O..
 8XX.X...............
 7.OX.X..........O...
 6O.X................
 5O....X..........O..  Left bottom is 5 libs vs 6 libs.
 4OOOOOX.........O...  White(O) must live
 3XXXXOX..........O..
 2XO.XOX.............  DCNN answer is White(O) lives 39%
 1.OOXOX.............  DCNN best move playout also fails.
  ABCDEFGHJKLMNOPQRST

(;GM[1]SZ[19]KM[0.5]RU[Chinese]AB[da][fa][ja][bb][cb][db][eb][fb][ib]
[jb][ic][dd][ed][fd][gd][hd][be][ce][af][al][bl][dl][cm][em][cn][fo]
[fp][aq][bq][cq][dq][fq][ar][dr][fr][ds][fs]AW[ba][ga][ia][ab][gb]
[bc][cc][dc][ec][fc][gc][oc][qc][ad][bd][pd][qf][qi][qk][bm][pm][an]
[ao][qo][ap][bp][cp][dp][ep][pp][eq][qq][br][er][bs][cs][es])

Problem1, B lives. 3 libs vs 5 libs. share 2 libs. (top)
Problem2, B lives. (right bottom)

(;GM[1]SZ[19]KM[0.5]TM[]RU[Chinese];B[pe];W[qp];B[dq];W[cd];B[op];
W[do];B[fq];W[cq];B[cp];W[dp];B[bq];W[cr];B[ep];W[dr];B[eq];W[co];
B[bp];W[dk];B[bo];W[bn];B[cn];W[dn];B[cm];W[dm];B[cl];W[dl];B[pm];
W[oo];B[po];W[pp];B[oq];W[no];B[qo];W[ro];B[rn];W[rp];B[qm];W[mq];
B[qr];W[rr];B[qq];W[rq];B[nr];W[mr];B[rs];W[pq];B[pr];W[qs];B[ps];
W[sr];B[ns];W[nm];B[so];W[pi];B[sp];W[ed];B[jd];W[ld];B[gd];W[lf];
B[nd];W[jf];B[ec];W[dc];B[fc];W[gf];B[he];W[ig];B[fn];W[fo];B[eo];
W[ck];B[bm];W[fm];B[gn];W[gm];B[hn];W[hm];B[in];W[im];B[jn];W[jq];
B[jm];W[hq];B[jl];W[en];B[gp];W[pg];B[qf];W[qc];B[oc];W[oe];B[pc];
W[pd];B[qd];W[od];B[mc];W[rd];B[qe];W[rc];B[rh];W[qg];B[rg];W[lc];
B[qb];W[rb];B[pb];W[re];B[rf];W[qk];B[ra];W[ne];B[se];W[md];B[nc];
W[jb];B[ic];W[ib];B[hb];W[hc];B[hd];W[ha];B[gb];W[ie];B[fe];W[id];
B[jc];W[kb];B[ff];W[fg];B[eg];W[dg];B[df];W[eh];B[de];W[dd];B[db];
W[cb];B[ef];W[cg];B[be];W[eb];B[fh];W[gg];B[dh];W[ch];B[ei];W[ci];
B[hi];W[ij];B[hj];W[fj];B[fi];W[ik];B[ii];W[ji];B[jh];W[ih];B[kh];
W[ki];B[lh];W[li];B[mh];W[gh];B[gi];W[kf];B[mi];W[jj];B[hf];W[if];
B[fk];W[hk];B[bf];W[bc];B[ej];W[bg];B[of];W[nf];B[og];W[mj];B[oh];
W[nj];B[mb];W[lb];B[la];W[pf];B[ph];W[gk];B[gj];W[ek];B[mg];W[mf];
B[ia];W[ge];B[fd];W[fb];B[ga];W[di];B[eh];W[hh];B[kd];W[le];B[kg];
W[ng];B[nh];W[qi];B[qh];W[ja];B[gc];W[bd];B[lg];W[lm];B[jg];W[ms];
B[ma];W[fl];B[ha];W[af];B[hc];W[ae];B[je];W[da];B[ka];W[ce];B[kc])

Problem3, W lives. 3 libs vs 4 libs. (top left)
Problem4, B dead.                    (right bottom)

(;GM[1]SZ[19]KM[7.5]RU[Chinese];B[qd];W[pp];B[dd];W[dp];B[qn];W[nq];
B[oc];W[pj];B[qh];W[qp];B[oi];W[oj];B[nj];W[nk];B[mk];W[pi];B[ph];W[ni];
B[mj];W[oh];B[lh];W[nl];B[ml];W[nm];B[ln];W[ri];B[lp];W[mo];B[mn];W[nn];
B[lo];W[dm];B[ic];W[cf];B[cd];W[ef];B[dg];W[df];B[dj];W[fd];B[fe];W[ee];
B[ed];W[fc];B[ge];W[db];B[cb];W[fb];B[gg];W[be];B[bd];W[dc];B[bb];W[cc];
B[bc];W[hd];B[hc];W[gd];B[if];W[id];B[jd];W[je];B[kd];W[ie];B[fg];W[ch];
B[ci];W[dh];B[ei];W[bi];B[bj];W[eh];B[bh];W[bg];B[ai];W[fh];B[eg];W[cg];
B[gi];W[gh];B[hh];W[hi];B[ih];W[fi];B[gj];W[fj];B[fk];W[gk];B[hj];W[ek];
B[fl];W[dk];B[ck];W[di];B[ke];W[jf];B[kf];W[jg];B[ig];W[hb];B[ib];W[gb];
B[ec];W[eb];B[ca];W[ad];B[jh];W[kg];B[lg];W[ac];B[ab];W[ae];B[fq];W[fp];
B[gp];W[fo];B[dq];W[eq];B[er];W[ep];B[gr];W[go];B[ho];W[hp];B[gq];W[hn];
B[io];W[hk];B[ij];W[cq];B[hl];W[in];B[mq];W[nr];B[ik];W[gm];B[gl];W[jo];
B[ip];W[jp];B[iq];W[jm];B[cm];W[cn];B[cl];W[em];B[og];W[ng];B[nf];W[rh];
B[bn];W[co];B[rk];W[rj];B[ro];W[rp];B[sp];W[ql];B[rq];W[qq];B[rr];W[qr];
B[bo];W[dr];B[rm];W[rl];B[rg];W[pg];B[of];W[qg];B[he];W[rf];B[gc];W[ce];
B[rs];W[sn];B[sm];W[so];B[oo];W[no];B[op];W[np];B[oq];W[or];B[da];W[ea];
B[sq];W[rd];B[qf];W[sg];B[pr];W[pq];B[po];W[ps];B[pf];W[qi];B[rc];W[qe];
B[sd];W[re];B[pd])

Problem5, B dead.  5 libs vs 6 libs   (top left)

(;GM[1]SZ[19]KM[0.5]AB[dd][pd][dp][pp];W[nq];B[pn];W[fq];B[dn];W[jp];
B[ci];W[fc];B[cf];W[nc];B[lc];W[ne];B[qf];W[kd];B[oc];W[kc];B[nb];W[mc];
B[mb];W[lb];B[nd];W[md];B[od];W[ql];B[ol];W[qi];B[ph];W[rn];B[pi];W[qj];
B[mo];W[lp];B[mp];W[mq];B[lo];W[kp];B[jn];W[fe];B[ec];W[fb];B[me];W[ld];
B[mf];W[cq];B[dq];W[dr];B[cp];W[br];B[bq];W[cr];B[gn];W[dj];B[cj];W[dk];
B[ck];W[dl];B[cl];W[eh];B[ef];W[ff];B[eg];W[fh];B[fg];W[hh];B[gg];W[ii];
B[gh];W[gi];B[hg];W[dh];B[di];W[fj];B[ei];W[fi];B[ej];W[fl];B[ek];W[hm];
B[fk];W[hk];B[gk];W[ik];B[hn];W[ig];B[if];W[jg];B[he];W[hc];B[ic];W[ib];
B[id];W[jb];B[gl];W[ch];B[pk];W[qh];B[qg];W[db];B[cc];W[bg];B[bf];W[bb];
B[cb];W[ca];B[eb];W[ea];B[bc];W[ee];B[de];W[dg];B[df];W[gf];B[hf];W[gd];
B[jf];W[kf];B[kg];W[je];B[jh];W[ih];B[kh];W[bi];B[jj];W[im];B[ij];W[hj];
B[ji];W[hi];B[hl];W[il];B[kl];W[kk];B[in];W[jk];B[lk];W[lj];B[mi];W[ll];
B[mk];W[km];B[jm];W[jl];B[kn];W[kj];B[mj];W[af];B[ae];W[bd];B[ba];W[aa];
B[ag];W[ah];B[mm];W[af];B[cg];W[ag];B[be];W[cd];B[ac];W[da];B[ab];W[ba];
B[dc];W[fa];B[ge];W[fd];B[hd];W[hb];B[lf];W[ke];B[gm];W[bj];B[bk];W[jc];
B[qo];W[ro];B[rp];W[ed])

Problem6, B lives. 6 libs vs 7 libs   (bottom left)

(;GM[1]SZ[19]KM[0.5]RU[Chinese];B[pd];W[dp];B[qp];W[dd];B[fq];W[cn];
B[kq];W[qj];B[pk];W[qk];B[ql];W[rl];B[pj];W[rm];B[qi];W[pl];B[pi];W[ol];
B[po];W[qm];B[mp];W[nc];B[nd];W[md];B[ne];W[oc];B[pc];W[me];B[ic];W[gc];
B[lb];W[lc];B[mb];W[mc];B[jc];W[od];B[oe];W[pe];B[pf];W[qe];B[qf];W[rd];
B[pb];W[of];B[nf];W[og];B[ng];W[oh];B[rf];W[nh];B[mg];W[kh];B[lh];W[li];
B[kg];W[mh];B[lg];W[jg];B[ke];W[kf];B[jh];W[lf];B[ki];W[ih];B[kj];W[jf];
B[mi];W[lj];B[lk];W[mj];B[ni];W[kk];B[ii];W[hi];B[ij];W[nj];B[oi];W[hj];
B[ik];W[kl];B[hk];W[nb];B[kc];W[hd];B[gb];W[fb];B[hb];W[gk];B[rc];W[gl];
B[ll];W[lm];B[nk];W[on];B[cf];W[ch];B[cd];W[cc];B[bc];W[bd];B[ce];W[bb];
B[dc];W[cb];B[ed];W[ec];B[de];W[ef];B[db];W[fa];B[ca];W[ba];B[da];W[fd];
B[ac];W[ad];B[ab];W[aa];B[ac];W[np];B[nq];W[op];B[oq];W[pq];B[pp];W[qq];
B[rq];W[rr];B[rp];W[ma];B[kb];W[or];B[nr];W[mo];B[pr];W[mm];B[mk];W[jn];
B[be];W[hq];B[dq];W[cq];B[ep];W[dr];B[er];W[eq];B[do];W[dq];B[co];W[bo];
B[fo];W[gn];B[bp];W[bn];B[cp];W[gp];B[fp];W[bq];B[gr];W[hr];B[ap];W[aq];
B[br];W[ar];B[bs];W[es];B[fr];W[ds];B[dn];W[cl];B[dm];W[dl];B[fs];W[en];
B[fn];W[em];B[fm];W[el];B[eo])

Problem7, B lives. big eye vs small eye

(;GM[1]SZ[19]KM[0.5]RU[Chinese]AB[gm][bo][co][do][eo][fo][go][bp][hp]
[bq][eq][fq][gq][hq][er][hr][bs][cs][es][hs]AW[ho][io][cp][dp][ep][fp]
[gp][ip][cq][iq][ar][br][cr][dr][fr][gr][ir][ds][gs][is])

Regards,
Hiroshi Yamashita

_______________________________________________
Computer-go mailing list
Computer-go <at> computer-go.org
http://computer-go.org/mailman/listinfo/computer-go
Rainer Rosenthal | 1 Feb 22:03 2016
Picon

Re: Mastering the Game of Go with Deep Neural Networks and Tree Search

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Robert: "Hey, AI, you should provide explanations!"
AI: "Why?"
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Cheers,
Rainer
> Date: Mon, 1 Feb 2016 08:15:12 -0600
> From: "Jim O'Flaherty" <jim.oflaherty.jr <at> gmail.com>
> To: computer-go <at> computer-go.org
> Subject: Re: [Computer-go] Mastering the Game of Go with Deep Neural
> 	Networks and Tree Search
> Message-ID:
> 	<CAKX5Gkjc7J0UQ_PMxyUmYFRe7r+7ydLTigBNa5Oo7KvnZq7qNA <at> mail.gmail.com>
> Content-Type: text/plain; charset="utf-8"
>
> Robert,
>
> I'm not seeing the ROI in attempting to map human idiosyncratic linguistic
> systems to/into a Go engine.

_______________________________________________
Computer-go mailing list
Computer-go <at> computer-go.org
http://computer-go.org/mailman/listinfo/computer-go

Gmane