Preset books and databases, with a 'hand-crafted' opening book of approximately 4,000 positions, an extended opening book with scored moves from a 700,000 game database, and an endgame database of pre-evaluated scenarios where six pieces or fewer remain on the board.Ĭomputational speed and scale with 30 IBM RS/6000 processor nodes and 480 special-purpose single-chip chess processors. Selective non-uniform (uneven) tree development, using heuristics to determine which nodes are worth expanding.Ī multi-factor evaluation function that considered 8,000 game features. The use of minimax searches with alpha-beta pruning, iterative deepening, and limited quiescence. Deep Blue’s Implementationĭeep Blue’s success came from a number of factors, including: The broad approach developed iteratively with each new generation, generally adding new methods and refining previous ones- as opposed to shifting to radically new approaches. The team that developed Deep Blue had worked on developing successive generations of chess systems since the 1980's with Deep Blue being the fifth generation. Deep Blue has since been retired to the Smithsonian Museum. Deep Blue defeated the then world chess champion Garry Kasparov in a 1997 exhibition match.
Backgroundĭeep Blue was a computer chess system developed at IBM Research in the mid-1990's in an effort to realise a "world-class chess machine". This review considers the project’s techniques and its results (largely as conveyed by the paper, with some additional commentary). I recently summarised the Deep Blue paper by Campbell, Hoane, and Hsu ( Artificial Intelligence, 2002) for a course that I'm doing.
0 kommentar(er)
