25,161
edits
Changes
Created page with "'''Home * People * Shu Yokoyama''' FILE:ShuYokoyama.jpg|border|right|thumb|link=https://acg2015.wordpress.com/videos-of-presentations/| Shu Yokoyama <ref>..."
'''[[Main Page|Home]] * [[People]] * Shu Yokoyama'''
[[FILE:ShuYokoyama.jpg|border|right|thumb|link=https://acg2015.wordpress.com/videos-of-presentations/| Shu Yokoyama <ref>Shu Yokoyama lecturing on ''Parameter-Free Tree Style Pipeline in Asynchronous Parallel Game-Tree Search''/ at [[Advances in Computer Games 14]], [https://acg2015.wordpress.com/videos-of-presentations/ Shu Yokoyama — slides]. Video Download, July 03, 2015, capture at 2:55</ref> ]]
'''Shu Yokoyama''',<br/>
a Japanese computer scientist affiliated with the Graduate School of Arts and Sciences, [https://en.wikipedia.org/wiki/University_of_Tokyo University of Tokyo]. His research interests include game playing algorithms and [[Search|search]], and in particular massively [[Parallel Search|parallel]] or distributed search algoritms. At the [[Advances in Computer Games 14]] conference in [[Leiden University|Leiden]] 2015, he introduced the distributed search algorithm dubbed '''P-GPP''', or pipelined game position parallelization.
<span id="PGPP"></span>
=P-GPP=
Along with [[Tomoyuki Kaneko]] and [[Tetsuro Tanaka]] <ref>[[Tetsuro Tanaka]], [[Tomoyuki Kaneko]] ('''2010'''). ''Massively Parallel Execution of Shogi Programs''. The Special Interest Group Technical Reports of IPSJ. 2, Vol. GI-24, No.8 (Japanese)</ref>, Shu Yokoyama worked on pipelined game position parallelization (P-GPP), applied to [[Shogi]] inside [[GPS Shogi]], and to [[Chess]] using [[Stockfish|Stockfish DD]]. GPP conducts a [[Minimax|minimax]] search by integrating the results obtained locally by workers assigned to [[Leaf Node|leaf nodes]] of a master [[Search Tree|tree]]. The [[Root|root]] of a master tree corresponds to the current [[Chess Position|game position]], and the number of nodes of the master tree must be the number of workers available. Neither [[Transposition Table|transposition table]] nor [[Window|windows]] are shared. P-GPP both extends [[GridChess#OptimisticPondering|optimistic pondering]] <ref>[[Kai Himstedt]] ('''2012'''). ''GridChess: Combining Optimistic Pondering with the Young Brothers Wait Concept''. [[ICGA Journal#35_2|ICGA Journal, Vol. 35, No. 2]]</ref>, and game position parallelization by improving worker management. In P-GPP, positions are assigned to workers automatically by a [https://en.wikipedia.org/wiki/Greedy_algorithm greedy algorithm] on the basis of [https://en.wikipedia.org/wiki/Realization_%28probability%29 realization] probabilities, acquired from game records and a playing program. The realization probability of a node, defined as the product of the transition probability of each move, is the probability that the corresponding sequence of moves is actually played. By definition, the realization probability of the root is one. After making a move during the course of the game, all workers formerly assigned to siblings of the made move were reassigned to new leaves of the growing new root, deepening and widening the search tree.
[[Stockfish|Stockfish DD]] was adopted as a worker program, adding a function of reporting information extending the [[UCI]] protocol. Each worker and the master are connected via standard [https://en.wikipedia.org/wiki/Transmission_Control_Protocol TCP] [https://en.wikipedia.org/wiki/Network_socket sockets]. The master is implemented in [[Cpp|C++]] with the [https://en.wikipedia.org/wiki/Boost_%28C%2B%2B_libraries%29 boost]/[https://en.wikipedia.org/wiki/Asio_C%2B%2B_library asio] library. For a worker, a utility program [https://en.wikipedia.org/wiki/Netcat Netcat] is adopted as a proxy connecting [https://en.wikipedia.org/wiki/Standard_streams standard streams] and a TCP socket. To simulate a distributed environment, they used at most 64 cores in two computers each of which is equipped with two [[Intel]] [[x86-64|Xeon E5-4650]] processors. Stockfish ran as a sequential program using a single [[Thread|thread]]. Each worker was allowed to use 32MiB for its transposition table . Shu Yokoyama et al confirmed improved playing strength with up to sixty Stockfish workers <ref>[[Shu Yokoyama]], [[Tomoyuki Kaneko]], [[Tetsuro Tanaka]] ('''2015'''). ''Parameter-Free Tree Style Pipeline in Asynchronous Parallel Game-Tree Search''. [[Advances in Computer Games 14]], [http://www.graco.c.u-tokyo.ac.jp/~kaneko/papers/acg2015-yokoyama.pdf pdf]</ref>.
=See also=
* [[GridChess]]
: [[GridChess#OptimisticPondering|Optimistic Pondering]]
* [[Stockfish]]
=Selected Publications=
* [[Shu Yokoyama]], [[Tomoyuki Kaneko]], [[Tetsuro Tanaka]] ('''2015'''). ''Parameter-Free Tree Style Pipeline in Asynchronous Parallel Game-Tree Search''. [[Advances in Computer Games 14]], [http://www.graco.c.u-tokyo.ac.jp/~kaneko/papers/acg2015-yokoyama.pdf pdf]
=External Links=
* [https://acg2015.wordpress.com/videos-of-presentations/ Shu Yokoyama — slides] from [[Advances in Computer Games 14]], Video Download, July 03, 2015
=References=
<references />
'''[[People|Up one level]]'''
[[FILE:ShuYokoyama.jpg|border|right|thumb|link=https://acg2015.wordpress.com/videos-of-presentations/| Shu Yokoyama <ref>Shu Yokoyama lecturing on ''Parameter-Free Tree Style Pipeline in Asynchronous Parallel Game-Tree Search''/ at [[Advances in Computer Games 14]], [https://acg2015.wordpress.com/videos-of-presentations/ Shu Yokoyama — slides]. Video Download, July 03, 2015, capture at 2:55</ref> ]]
'''Shu Yokoyama''',<br/>
a Japanese computer scientist affiliated with the Graduate School of Arts and Sciences, [https://en.wikipedia.org/wiki/University_of_Tokyo University of Tokyo]. His research interests include game playing algorithms and [[Search|search]], and in particular massively [[Parallel Search|parallel]] or distributed search algoritms. At the [[Advances in Computer Games 14]] conference in [[Leiden University|Leiden]] 2015, he introduced the distributed search algorithm dubbed '''P-GPP''', or pipelined game position parallelization.
<span id="PGPP"></span>
=P-GPP=
Along with [[Tomoyuki Kaneko]] and [[Tetsuro Tanaka]] <ref>[[Tetsuro Tanaka]], [[Tomoyuki Kaneko]] ('''2010'''). ''Massively Parallel Execution of Shogi Programs''. The Special Interest Group Technical Reports of IPSJ. 2, Vol. GI-24, No.8 (Japanese)</ref>, Shu Yokoyama worked on pipelined game position parallelization (P-GPP), applied to [[Shogi]] inside [[GPS Shogi]], and to [[Chess]] using [[Stockfish|Stockfish DD]]. GPP conducts a [[Minimax|minimax]] search by integrating the results obtained locally by workers assigned to [[Leaf Node|leaf nodes]] of a master [[Search Tree|tree]]. The [[Root|root]] of a master tree corresponds to the current [[Chess Position|game position]], and the number of nodes of the master tree must be the number of workers available. Neither [[Transposition Table|transposition table]] nor [[Window|windows]] are shared. P-GPP both extends [[GridChess#OptimisticPondering|optimistic pondering]] <ref>[[Kai Himstedt]] ('''2012'''). ''GridChess: Combining Optimistic Pondering with the Young Brothers Wait Concept''. [[ICGA Journal#35_2|ICGA Journal, Vol. 35, No. 2]]</ref>, and game position parallelization by improving worker management. In P-GPP, positions are assigned to workers automatically by a [https://en.wikipedia.org/wiki/Greedy_algorithm greedy algorithm] on the basis of [https://en.wikipedia.org/wiki/Realization_%28probability%29 realization] probabilities, acquired from game records and a playing program. The realization probability of a node, defined as the product of the transition probability of each move, is the probability that the corresponding sequence of moves is actually played. By definition, the realization probability of the root is one. After making a move during the course of the game, all workers formerly assigned to siblings of the made move were reassigned to new leaves of the growing new root, deepening and widening the search tree.
[[Stockfish|Stockfish DD]] was adopted as a worker program, adding a function of reporting information extending the [[UCI]] protocol. Each worker and the master are connected via standard [https://en.wikipedia.org/wiki/Transmission_Control_Protocol TCP] [https://en.wikipedia.org/wiki/Network_socket sockets]. The master is implemented in [[Cpp|C++]] with the [https://en.wikipedia.org/wiki/Boost_%28C%2B%2B_libraries%29 boost]/[https://en.wikipedia.org/wiki/Asio_C%2B%2B_library asio] library. For a worker, a utility program [https://en.wikipedia.org/wiki/Netcat Netcat] is adopted as a proxy connecting [https://en.wikipedia.org/wiki/Standard_streams standard streams] and a TCP socket. To simulate a distributed environment, they used at most 64 cores in two computers each of which is equipped with two [[Intel]] [[x86-64|Xeon E5-4650]] processors. Stockfish ran as a sequential program using a single [[Thread|thread]]. Each worker was allowed to use 32MiB for its transposition table . Shu Yokoyama et al confirmed improved playing strength with up to sixty Stockfish workers <ref>[[Shu Yokoyama]], [[Tomoyuki Kaneko]], [[Tetsuro Tanaka]] ('''2015'''). ''Parameter-Free Tree Style Pipeline in Asynchronous Parallel Game-Tree Search''. [[Advances in Computer Games 14]], [http://www.graco.c.u-tokyo.ac.jp/~kaneko/papers/acg2015-yokoyama.pdf pdf]</ref>.
=See also=
* [[GridChess]]
: [[GridChess#OptimisticPondering|Optimistic Pondering]]
* [[Stockfish]]
=Selected Publications=
* [[Shu Yokoyama]], [[Tomoyuki Kaneko]], [[Tetsuro Tanaka]] ('''2015'''). ''Parameter-Free Tree Style Pipeline in Asynchronous Parallel Game-Tree Search''. [[Advances in Computer Games 14]], [http://www.graco.c.u-tokyo.ac.jp/~kaneko/papers/acg2015-yokoyama.pdf pdf]
=External Links=
* [https://acg2015.wordpress.com/videos-of-presentations/ Shu Yokoyama — slides] from [[Advances in Computer Games 14]], Video Download, July 03, 2015
=References=
<references />
'''[[People|Up one level]]'''
