Computer-Go am HPI

Ideensammlung zum Computer-Go-Seminar des HPI

ToDos

• Ausarbeitung der Gruppenaufteilung
  • verschiedene Themen der Gruppen?
  • konkurrierende Gruppen?
• Anforderungen an die Teilnehmer definieren
• Hardware-Anforderungen definieren
• Literatur zusammentragen
• Basic-Player evaluieren
• SVN organisieren
• Mailingliste einrichten
• Kooperation mit anderen Lehrstühlen klären
• sourceforge projekt? -> open source?

Eckpunkte

Teilnehmer

• 6 - 10 Teilnehmer
• Interessenten:
  • Marius Schneider
  • Tobias Pfeiffer
  • Martin Büttner
  • Martin Köppelmann
  • Lars Wassermann
  • Maximilian Schneider?
  • Maximilian Möller?
  • Henning Klein

mögliche Vertiefungsgebiete

• Monte Carlo Tree Search & co.
  • dynamisches Komi
• Parallelisierung (Thread / Cluster)
• Tsume Go
• direction of play
• statische Evaluationsfunktionen
• Kombination von verschiedenen Ansätzen (MCTS, direction of play, statische Evaluation...)
• Parameter Tuning
• Evaluieren des Players (menschliche / computer- Gegner)

Literatur

(Auszug aus den Veröffentlichungen der letzten Jahre)

Monte Carlo Tree Search

• PHD: A CONTRIBUTION TO REINFORCEMENT LEARNING; APPLICATION TO COMPUTER-GO (SYLVAIN GELLY)
• Combining expert, offline, transient and online knowledge in Monte-Carlo exploration (Chaslot et.al)
• PROGRESSIVE STRATEGIES FOR MONTE-CARLO TREE SEARCH (Chaslot et.al)
• CROSS-ENTROPY FOR MONTE-CARLO TREE SEARCH (Chaslot et.al)
• Combining Final Score with Winning Percentage by Sigmoid Function in Monte-Carlo Simulations (SHIBAHARA, KOTANI)
• On the Laziness of Monte-Carlo Game Tree Search in Non-tight Situations (Ingo Althofer)
• Transpositions and Move Groups in Monte Carlo Tree Search (Benjamin E. Childs et.al)
• Monte Carlo Temperature Discovery Search for Combinatorial Games (Philipp Hennig)
• Efficient Selectivity and Backup Operators in Monte-Carlo Tree Search (Rémi Coulom, Crazy Stones)
• Computing Elo Ratings of Move Patterns in the Game of Go (Rémi Coulom, Crazy Stones)
• Current Approaches in Computer Go (Slides) (Petr Baudis)

Parallelisierung

• Parallel Monte-Carlo Tree Search (Chaslot et.al)
• Parallel Monte-Carlo Tree Search with Simulation Servers (KATO, TAKEUCHI)
• THE PARALLELIZATION OF MONTE-CARLO PLANNING (S. Gelly et.al)

Other

• PHD: AI techniques for the game of Go (2004) (Erik van der Werf)
• Modelling Uncertainty in the Game of Go (David H. Stern et.al)
• Learning Position Evaluation for Go with Internal Symmetry Networks (Neuronale Netze) (Alan Blair)
• Coevolving Partial Strategies for the Game of Go (Genetic algorithm) (lokale Kopie)
• Lambda Depth-first Proof Number Search and its Application to Go (Yoshizoe, Kishimoto, M. Müller)
• MSc Thesis: Learning Patterns in the Game of Go (Emil H.J. Nijhuis)

Basicplayer

• Java
  • plug-and-go
  • Java-Bot by Don Dailey
• C++
  • fuego
• Go
  • Gongo

Links

online spielen

• Computer Go Server
• Connecting a Go-playing program to KGS
• Computer Go Tournaments on KGS

Community

• Go-Mailingliste
• DGoB-Forum Computer-Sektion

andere Player

• kifoo (FU Berlin)
• mogo (Uni Paris)
• fuego (Kanada Alberta)
• gnugo
• crazy stones
• NeuroGo (University of California)
• GoTools (Thomas Wolf) (Tsume Go)
• PD Dr. rer. nat. Ulf Lorenz (Darmstadt) (kein Player)

something else

• why programmers should play go
• SGF-Format Spezifikation
• Shodan Go Bet