andre/UoL
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Initials notes for the article

Browse Source
This commit is contained in:
levdoescode
2022-11-09 01:06:52 -05:00
parent 1aff4e9b30
commit eef5d04376
1 changed files with 18 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
CM3020 Artificial Intelligence/Week 1/2. A gamut of games (2001) notes.md Normal file
View File

@ -0,0 +1,18 @@
Games are a good starting point for AI because the rules are defined and the number of ways players can interact is limited.
1951 Alan Turing did a hand simulation of his comptuer chess algorithm because the resources were not available to program it. The algorithm lost to a weak player.
In the next 50 years these game problems were solved by advancements in hardward, better understanding of the problems at hand and the algorithms being employed.
The alpha-beta search algorithm has been the biggest contributor in the advancement of game playing AI. It took a central stage in the hay day of chess AI.
Some enchances of alpha beta search are iterative deepening, caching previously seen subtree results (transposition tables), successor reordering, search extensions and reductions, probabilistic cutoffs, parallel search.
At the heart of game-playing programs there is an evaluation function. At the beginning of AI research, Heuristic knowledge combined with deep search worked better than trying to immitate human cognitive processes.
Two new techniques (2001) at the forefront of games research are:
* Monte Carlo simulation
* Temporal-difference learning
In Monte Carlo simulation is used in nondeterministic games by providing a representative sample to product a statistical profile of the desired outcome. Used successfully in bridge, poker, and Scrabble.