44

You have little chance in your dissertation of surpassing state of the art chess engines. Perhaps you could find a hook which hasn't been explored so much. One idea is to train your program to play amateur chess in a convincing way. Can a program pass a sort of chess Turing Test where a human-player couldn't tell if they were playing an AI or playing a ...


17

You might be interested in AlphaZero and its derivatives. AlphaZero is the original neural-network based chess engine; since then there have been various other attempts (Leela Chess Zero, AllieStein, the recent Fat Fritz ...) to replicate AlphaZero's ideas. The key paper to read is this one. The data right now seems to indicate that although these neural ...


11

To be frank, I was a little surprised that this was a draw per the computer from move one all the way through to your stalemate. You did not make a mistake. My initial thought was to play the black king to g8-f8-e7, but the problem is that white can keep the K out of e7 with his own king, or keep you on the back rank, or force you to make pawn moves that are ...


11

I had an extra pawn and was closer to my pawns then White It seems you simply overestimated these two factors. An extra pawn alone is not a decisive advantage, it matters a lot where it is located. If the f7 pawn was on a7, you'd win. The kings are actually positioned to your disadvantage, White's is nicely centralized and yours is stuck in the corner. ...


10

There are plenty of possible Phd-topics that connect chess and machine learning. The more interesing and feasible ones have nothing to do with building better engines. Here are some ideas: Human players have a particular playing style. Is it possible to learn to extract some metric of style from games that allows to assign games to players with a certain ...


8

Losing an exchange is itself enough to make the move weak. You go from being up R+P for N, which is a winning advantage, to up just one pawn with opposite color Bishops, which could often be drawn. The move is also weak because you let White off the hook: you were on the attack against White's exposed King, so the trade of Queens made the position "easier ...


8

I have been looking at this with a fairly strong computer for about 15 minutes (and the judgment of having been a Master for 30+ years and forcing in plausible lines), and I am convinced that this is a positional draw. White is obviously worse per the computer eval, but he can keep the position exactly as it is by either moving his K from e2 to d3 and back (...


7

Well, I am amazed. I thought that it would be another case of a free online engine just not being that strong, but my Stockfish, and Deep Rybka, both like g4 there. I know I would never have played that move in a million years since the eval is very close to other more normal moves like Re1, but here is what it sees. It thinks that you can get in g5 when ...


6

A couple of ideas: Does chess help kids do better in school? That question has long been debated. Do certain businesses with chess players, especially strong chess players, have an advantage in certain types of business, like finance?


6

I am not a computer scientist. But I have been a chess computer enthusiast since the 1980's. My first engine was a tabletop Novag in about 1988. I read David Levy's book "how to get the most out of your chess computer". The two traditional weaknesses in chess engines have always been [1] The horizon effect. David Levy describes it as being like a man in a ...


6

And one last one from a real game. Black to move and win. [Date "2003.??.??"] [White "Ibrahim"] [Black "Ibarra"] [FEN "3q4/1p1n1pk1/p1r1p1p1/2P1Pn1p/1P1N1P2/B5rP/5RPK/Q3R3 b - - 0 1"] 1... Nxd4 $1 2. Kxg3 (2. Qxd4 Rxa3 3. Rd1 Qh4 {is just up a piece.}) 2... Qh4+ $3 3. Kxh4 Nf5+ 4. Kg5 Nf8 5. g4 {and there is no defense to....} Nh7#


6

I agree that f5 is irrelevant. A weakness is only a weakness if it can be exploited. The main reason that g5 works here is that it can force the N away from f3, and then your N can go to d4 with not only a direct threat of Nxc2, which is VERY inconvenient to defend, and the more decisive threat of Nf3!. These threats are so strong that the computer thinks ...


5

Playing ...d5 isn't your typical Pirc idea, although it can be played under the right circumstances. In the diagram, White's d3-bishop and f3-knight are vulnerable to a potential ...e4 push. In addition, your g4-bishop provides some tactical possibilities by pinning the knight. Concretely speaking, everything just works nicely for an immediate challenge in ...


5

In most early-stage Pircs, black is slightly worse, so of course, the computer will prefer any move that equalizes, or in this case, gives you a slight advantage. White is not well-placed in the center due to the pin on the Nf3, and the Bd3 interferes with the Qd1's protection of the center (d4) too. Black can effectively "blow-up" the center here, which ...


5

The solution to that is much easier than it looks, and a fairly common type of position in practical play. After 19.Nd5! taking the nice outpost square, if black responds with 19…Nd5 then white can simply play 20.Bd5 Qb4, then Qxa6 just leaves black positionally lost despite even material. In positions like that where you have a rook behind the a-pawn, and ...


5

What endgame principles are there/exist that would help? You need to learn to recognise which pawn formations allow you to force a passed pawn and which don't. The pawn formation on the kingside is almost identical to the queenside pawn formation in the Berlin Defence which is known to be unable to force a passed pawn. No passed pawn = no win.


5

The problem with that move is a number of things. You gave up the bishop pair. His attack on your king is stronger than yours on his king, and you traded off a defender. You have no time to go after his king since yours is threatened with mate on the move. Already with three minors traded, his threats aside, there might not be enough firepower to attack his ...


5

A missed variation from the game Najer-Nepomniachi, Moscow 2006. Black to move and win. [FEN "5rk1/3R1p2/4p2p/4P1pP/2q4r/P1n1BQ2/2P2PP1/4R1K1 b - - 0 1"] 1... g4 2. Qg3 Qf1+ $3 3. Rxf1 (3. Kxf1 Rh1#) 3... Ne2#


5

You are correct that 18.Bd6? is, indeed, not the best move (the simple Re1! threatens Nc7-a8 winning on material). Bd6 has widely been considered a mistake for many years now. Black holds with 18...Qxa1+ 19.Ke2 Qb2! 20.Bc5 Qc2 21.Kf1 Qc5 22.Qf4 and black has several moves that are equal, but in the last few moves, it is white that has to find the only moves ...


4

I think there are a few things here. When playing a gambit line that maybe you do not know, the first thing I would recommend is not necessarily taking it. Either 3.Ne5 or even 3.d4 look safer. On move four, you had the choice of taking two pawns, one that developed his N, and the other, which took a center pawn. The center P was easily the right way to go ...


4

This was played almost at Fischer’s height of strength, so he almost surely would have won anyway after 66.Ke3 also. I think that white thought two things by playing 66.ba. First, that it reduced the number of pawns, which is almost always good when you are in a worse, or losing, endgame. Second, that the bishop might be able to defend a2 along the a2-g8 ...


4

White to win. L Kubbel 1927 [FEN "8/8/1p6/kp3p2/p7/1PP1p2B/5p2/5R1K w - - 0 1"] 1. Ra1 $1 {Threatening Bf1 stopping the pawns.} e2 2. Bf1 $1 e1=Q 3. Rxa4+! bxa4 4. b4#


3

White has a wealth of factors working to his advantage: A lead in development. A massive centre. Having control over the dark squares. These factors allow him to get away with pushing g4. Sure, it weakens the f4- and h4-squares, but his dark-squared bishop can defend them (and Black lacks a dark-squared bishop of his own). There's also the fact that the h6-...


3

Feasibility of non-search based approaches I'm going to talk about the feasibility of using a non-search based approach. Immediately, some questions that come to mind: Does there exist a good static_eval function that takes a (Position, PlayerTurn) and returns some Score? ¹ Small changes in positions can often lead to massive differences in evaluation. ...


3

Have you considered something like general game play? If you can train an AI to be good at both chess and other board based games (with the same network) you could show some interesting results. This paper explores that concept for MCTS This article explores the concept in RL.


3

I think a cool project which mixes chess and machine learning is to build a chess bot for 4 players chess. Such variant can be played on chess.com and if I remember correctly, they don't have particularly strong bots at the moment. Probably the most fun would be to use reinforcement learning and let the bot learn by playing against human players on the ...


3

Yes, it is really an easy win. The first move is 1...Qe4; 2.Kg4 Kc7, and note that white cannot make any progress without losing one of the pawns, leaving an easy win since your Q can get to g8 after winning the first pawn. [FEN "qk6/8/8/6P1/5P2/6K1/8/8 b - - 0 1"] 1... Qe4 2. Kg4 Kc7 This is with it going the other way. It is harder, no question, but ...


2

I simply do not agree with your computer there. It is probably a bit of the "horizon effect", which is when a computer plays a bad move because it puts the refutation beyond its ability to calculate, or "horizon". White played this opening every passively, and really should be looking to break up black's center so he is not smothered, and b4 does not seem ...


2

After 1...Nxd5 2.Bxd5 Qxb4 3.Qxa6 White has a clear advantage. He has two very nice bishops and an outside passed pawn on the a-file. Black's d6-pawn is a clear weakness, his bishop on d8 lacks prospects, and the list goes on. However, Stockfish actually gives 1...Nxd5 as its top choice, with 1...Nc7 coming in as a close second. In either case though, it ...


2

In this situation, as a defender needing a draw, you just need to ask yourself: "Can I keep the pawns doubled?" Or, as an attacker trying to win,: "Can I undouble my pawns?" The "yes" answer will mean you should be able to succeed in the respective task.


Only top voted, non community-wiki answers of a minimum length are eligible