Othello

1
Othello

• Sean Farrell
• June 18, 2009

2
Othello

• Two-player game played on 8x8 board
• All pieces have one white side and one black side
• Initial board setup is shown right, with valid
  moves marked as dots

3
The Evolution of Strong Othello ProgramsMichael
Buro, 2003

• First computer Othello tournaments in 1979
• In 1980, first time a World-champion lost a game
  of skill against a computer
• 6-0 defeat of then human World-champion in 1997

4
Evaluation Function Evolution

• Mini-max search used to estimate the chance of
  winning for the player with current move
• Construct function using features of board state
  that correlate with winning
• Important features
• Disc stability stable discs cannot be flipped
• Disc mobility move options
• Disc parity last move opportunities for every
  empty board region

5
IAGO (1982)

• Classic, hand-crafted evaluation function
• Features were chosen based on analysis of Othello
• Edge Stability Internal Stability
• Current Mobility Potential Mobility
• Evaluation parameters were done manually

6
BILL (1990)

• Partly pattern based, feature weights are learned
• Edge Stability
• Current Mobility Potential Mobility
• Sequence Penalty
• Evaluation speed increased by using pre-computed
  tables
• In comparison, BILL wins all games against IAGO,
  with 20 the thinking time

7
Logistello-1(1994)

• Pattern values learned independently from sample
  data
• Uses logistic regression to combine features
• Current Mobility Potential Mobility
• Patterns

8
Logistello-1(1994)

• Evaluation function is entirely table based
• All evaluation parameters are learned from sample
  positions
• Logistello uses selective search heuristic called
  ProbCut
• Dominated computer Othello until 1996

9
Logistello-2(1997)

• Joint learning of pattern values
• Assigning pattern values independently neglected
  pattern correlations
• Can assign arbitrary values to patterns whose
  meaning is not bound by limited human
  understanding of the problem
• Strength increase from 1994 version is ten-fold

10
Other Important Improvements

• Opening books
• Saves time
• Avoids falling into known strategic traps
• Less chance of losing two games in the same way
• End game search
• Allows for optimal play near end

11
Takeshi Murakami vs. LogistelloMichael Buro, 1997

• Both humans and computers have improved playing
  Othello
• Against imperfect human players it pays off to
  complicate endgame positions, not so with strong
  Othello programs

12
Human vs. Computer
Takeshi Murakami
Michael Buro and is program Logistello
13
Discovering Complex Othello StrategiesD. E.
Moriarty Risto Miikkulainen, 1997

• Found that experts search selective paths through
  pattern recognition
• Most programs used deeper searches to be
  effective
• Wanted a human-like approach

14
Neural Networks

• Networks learned Othello without previous
  knowledge
• No hand-coded rules or heuristics
• Strategies evolved through play
• No search mechanism
• Goal was to discover strategies

15
Implementation

• Neural networks relied on pattern recognition
• Used marker-based scheme with genetic algorithms
• Network architecture and weights evolve
• Population of 50 networks
• Initially evolved against random move maker

16
Results

• Positional strategy against random mover after
  100 generations

17
Results

• Mobility strategy against searcher after 2000
  generations

18
Articles

• M. Buro, The Evolution of Strong Othello
  Programs, 2003
• M. Buro, Takeshi Murakami vs. Logistello, 1997
• D. E. Moriarty R. Miikkulainen, Discovering
  Complex Othello Strategies Through Evolutionary
  Neural Networks, 1995

19
Websites

• Michael Buro http//www.cs.ualberta.ca/mburo/
• Neural Networks Research Group http//nn.cs.utexas
  .edu/
• Free online Othello game http//brittany.angloinfo
  .com/games/iago.asp

20
Questions ??