|
Post by bleh on Feb 13, 2011 16:45:49 GMT -5
1. can you expain monte carlo algorithim in more deatail
2. how do strings in kif format look like
|
|
|
Post by Jón Þór Kristinsson on Feb 13, 2011 17:54:03 GMT -5
Looking at the results from the GGP Competition in 2010 what did Ary and Maligne improve to get a better score then the CadiaPlayer that year?
Can the CadiaPlayer be tailored against any specific player when playing (Such as using some strategy that works well against Ary but might not work as well against any other Player)?
|
|
|
Post by gunnar on Feb 13, 2011 18:06:06 GMT -5
What are the biggest changes made to the CadiaPlayer in the last 1-2 years?
Is there a method not used by any of the agents that might make a strong agent?
|
|
|
Post by Elín Carstens on Feb 13, 2011 18:13:38 GMT -5
1) It is mentioned in the article that the work of automatically finding informed heuristics by problem relaxation is still a work in progress. Have there been any developments in this area?
2) In games where game-specific knowledge is required, what is the best way to apply it to GGP?
|
|
una
New Member
Posts: 12
|
Post by una on Feb 13, 2011 18:17:02 GMT -5
Does a GGP ever reach a state at which it has nothing more to learn? ( depends on the game maybe ?) What games are usually played in the GGP competition?
|
|
|
Post by gudrunht on Feb 13, 2011 18:23:50 GMT -5
Is the the CadiaPlayer limited to playing games like puzzles,cards and board-like games (like chess or poker) or can it handle some more complex computer games? What is the most advanced game the CadiaPlayer has succeeded in learning? And in that regard, where are it´s limits?
Have you worked on combining simulations and traditional game-tree serach as you mention at the end of the paper? How are you advancing the CadiaPlayer for it to reach it´s next level? What are your biggest obstacles for advancement?
What changed for the CadiaPlayer in the GGP competitions after 2008? Are we(HR) going to claim back "our" title this year?
P.S. What is a computer Go? (It´s mentioned a couple of times but I´m still unsure of what that is.)
|
|
|
Post by baldurb09 on Feb 13, 2011 18:26:16 GMT -5
1. What have the most important developments regarding General Game Playing been for the past decade? What advancements in General Game Playing and Artificial Intelligence have there been made independently of the increase in processing capabilities of computers.
2. The UCT Game Tree algorithm is used to keep track of the average return of simulations—could the aggregated metainformation of billions of moves (pertaining to a position in time, types of moves; in chess: “rook moved”, “promotion”, “castling” etc.) at a higher level be used to find some correlation for actions that are generally better than others and use that to help improve decision making?
|
|
|
Post by sindrib on Feb 13, 2011 18:28:22 GMT -5
How is the action-value function (for opponent modelling) adjusted between different game types?
What other fields (problems) might the Cadia-player be adapted to as a useful tool?
|
|
|
Post by gunnar on Feb 13, 2011 18:46:17 GMT -5
|
|
|
Post by niccolo on Feb 13, 2011 21:54:11 GMT -5
Is it correct to think that if a program can learn strategies for a game it can also learn strategies to solve real-world problems? What about the game specification then: is there any chance to get them evaluated automatically?
Your contribution was innovative for GGP. Concerning current and future works, are you fully satisfied with this approach (Monte Carlo) and then you focus in get it better or do you have also other innovative ides?
If I have to implement an agent for a game, will a GGP be better than a game specific AI?
|
|