Do any of you happen to know of any good journal papers regarding imperfect information games (like poker)? I just ordered The Mathematics of Poker, and plan to read that this week, but I'm looking for some more in depth information, as well. I'm not too versed in the rigors of game theory or anything, and I'm only vaguely familiar with Nash equilibrium, but I have a very solid math background and feel extremely comfortable with the basics I've been reading on wikipedia.

In the reviews for the aforementioned book, I read that most of the examples given are fundamental examples where basic decisions are allowed: bet or fold. It was also mentioned that there isn't enough computing power readily available to generate consistent +EV decisions for full (10 player) ring games. Is this true?

Lots of discussion has been devoted to the topic of whether it's even worth it to analyze poker with the principles of game theory, and while a perfect solution to the problem could result in bots being generated that are impossible to beat, I still think that's years away from occurring. Basically, I'm looking for another way to look at situations where the decision between a bet and a raise or a raise and a reraise is hazy at best.

look for andrew gilpin in google... he has written a couple of papers about computational game theory on poker

then the ferguson brothers

look for darse billings too

basicly you start with gilpin, look up every reference and every reference in the references and you will have everything ever written

you need access to all those online librarys though, beeing member of an university helps...

There are two main(correlated) problems with simulating poker:

First, you can't enumerate all possible plays as in chess and many other games
(there are infinite possible bet relations and player statistics in poker).
Modern chess programs are using a lot of tricks like a lookup tables
for early game strategies and excluding obviously bad moves,
but it's still an enumeration of all possible moves.

Even under these restrictions, a super computer like deep blue
is needed to beat the best players in the world. And i think that
even deep blue is far away from optimal play.

Poker needs a different approach with a higher complexity.

Second, many aspects of poker are related to pattern recognition,
mainly classifying your opponents. Pattern recognition is a
notoriously hard computational problem, it uses statistical analysis,
fuzzy logic and hierarchical decision processes.
Massive parallel pattern recognition is one of the remaining fields
in which humans outclass computers nowadays.

For example, Go is like poker one of the last games where
humans are way better, and it's based on pattern recognition, too.

If i remember correctly, even NL HU is not solved yet.

http://www.math.ucla.edu/~tom/papers/poker1.pdf
http://www.math.ucla.edu/~tom/papers/poker2.pdf
http://www.math.ucla.edu/~tom/papers/poker3.pdf
http://swansonsite.com/W/instructional/game_theory.pdf
http://staff.science.uva.nl/~faolieh...5MScThesis.pdf

well they are pretty close on solving HU games...

multiplayergames are kind of a hassle because of implicit collusion and so forth plus gametree gets bigger

but that is still the better approach than opponent modeling pattern recognition and the other stuff that computers are really bad at and usually not better than their builder...

the approach to optimal play is the only approach that has a realistic chance at beating "every" human beeing in poker.

JocK:
Very nice links, tyvm.

thanks for the links everyone; these'll be good reads while i'm stuck in class this week