1. Training Neural networks to play checkers
Daniel Boonzaaier
Supervisor – Adiel Ismail
September 2017

2. Contents Recap Project Goal Neural Network Particle Swarm Optimization
Training Results
System Testing
User Testing
Future Work
References
Project Demo
Questions

3. Recap Checkers Neural Network Particle Swarm Optimization
Creating an intelligent game playing agent using neural networks and particle swarm optimisation for the game of checkers.
Personal Best
Global Best

4. Project Goal
Determine the ability of Neural Networks to learn the game of Checkers from zero knowledge with a single depth look ahead.

5. Neural Network Score i.e. 0.67
The Neural Network is used to constantly evaluate the board and do calculations to determine the best move.
Every turn all possible moves are evaluated and scored.
The move with the highest score is chosen.
Calculation is done after a copy of the board vector has been made, a move made to the copy and then evaluated. This is done for every possible move.
Score i.e. 0.67

6. Particle Swarm Optimisation
Used for Training only.
Updates the weights of the Neural Networks based on a fitness calculated for each and a velocity function.
Fitness calculated by having Neural Networks play each other. Add 1 for a win. Minus 2 for a loss. Draw adds 0.
Particle = Weights of a Neural Network Agent
Fitness = 2
Fitness = -2
Fitness = 5

7. Training Results
Benchmark – Games played with randomly selected moves.
White win – 40,00%
Black win – 41,09%
Draws – 18,91%
Average – 40,54%

8. System Testing
Testing that all components work together correctly and remove any errors/bugs in code.

9. User Testing User testing for GUI.
Neural Network playing ability VS human players.
User feedback on the GUI, advice, bugs and other comments.

10. User Testing

11. User Testing

12. Future Work
Improvements can be made in training by increasing the depth of the look ahead.
Allow the Neural Network to evaluate future moves and consequences.
Single depth

13. Project Plan Term 1 – Research  Term 2 – Designing and Prototyping 
Research related works and determine requirements
Term 2 – Designing and Prototyping 
Design program and neural net
Term 3 – Implementation of Design 
Create program and train neural net
Term 4 – Testing and Refining 
Test competency and refine program if necessary

14. References
A. L. Samuel, “Some Studies in Machine Learning Using the Game of Checkers”, IBM Journal, Vol 3, No. 3, (July 1959)
K. Chellapilla, D. B. Fogel, “Evolving Neural Networks to Play Checkers Without Relying on Expert Knowledge”, IEEE Transactions on Neural Networks, Vol. 10, No 6, (Nov 1999)
N. Franken, A. P. Engelbrecht, “Evolving intelligent game-playing agents”, Proceedings of SAICSIT, Pages , (2003)
N. Franken, A. P. Engelbrecht, “Comparing PSO structures to learn the game of checkers from zero knowledge”, The 2003 Congress on Evolutionary Computation. (2003)
A. Singh, K. Deep, “Use of Evolutionary Algorithms to Play the Game of Checkers: Historical Developments, Challenges and Future Prospects”, Proceedings of the Third International Conference on Soft Computing for Problem Solving, Advances in Intelligent Systems and Computing 259, (2014)
H. Kwasnicka, A. Spirydowicz, “Checkers: TD (λ) Learning Applied for Deterministic Game”, Department of Computer Science, Wroclaw University of Technology, Poland. (June 2014)

15. Project Demo

16. Questions?