You will build upon the results of the Blondie24 book in groups of 2 or 3. Pick at least one "defect" from the Blondie24 process (e.g. initial population too small, too few generations, too few nodes/layers in neural network, ...) and compare the original to your "improved" version. You must create a graphical interface that allows you to watch any game played by your neural network. (This will also serve as a useful debugging tool ...) You must collaborate with the other groups so you can participate in a tournament.

Group selection: Submit a group name and a list of the members in your group by Thursday Feb 11.

Proposal & Presentation: (10%) Each group will give a 10 minute presentation describing their proposed project on Thursday Feb 25. You need to provide the following information:

• Who is part of this project?
• What are your goals?
• How are you going to measure your goals?
• Are your goals feasible? (i.e. computer resources, programming effort)
• What resources do you have? (i.e. computers, programs, experts, web, on-line game sites, ...)
• Show a spreadsheet the verifies the computation you need to do
• How are you going to validate the results? (i.e. on-line playing, good player, good program, ...)
• What will you have done at each 2 week interval?

Turn in a 2-3 page written summary. This will form your initial proposal and will need to be approved by me.

Progress Reports: (15%) On Thursday Mar 18, Thursday Apr 1 and Thursday Apr 15, you will have 10-15 minutes to show what you have done and what is left to do. Demos/stats/examples are the best way to show this. You will be graded on how well you convince me that you are on schedule. You will lose points if you fall behind. Key metrics include board evaluations per second, average time to play a game, and average time per generation. Any changes to your proposal need to be written and approved by me.

Final Presentation: (15%) You will have 15-20 minutes for a final presentation on Thursday Apr 29.

Final Report: (50%) Turn in a 5-10 page written report by Thursday May 6.

Tournaments: (10% total) Each group will participate in a tournament pitting their best neural network against each other group. There will be a dry-run on Thursday Apr 22 to work out any problems. Next, each group will play all of the other groups a maximum of 10 times (5 as white and 5 as black) and receive +1 for a win, 0 for a draw and -1 for a loss. Groups are encouraged to play as many of these games off-line as possible and report the results (including the game log) to the instructor. We will spend Tuesday May 4 playing games in class, starting with the groups that have played the fewest games. The point totals from the round robin will be worth 5% and determine the seeds for the double elimination tournament held on Thursday May 6. The results of the double elimination tournament will be worth 5%. If your neural net is unable to play on this day, you will receive no credit for the tournament.

Gameplay rules:

• There is a soft limit of 1 minute per move. This will be enforced by a protest from the opponent or instructor.
• Illegal moves. When your program receives a move from an opponent, it may declare the move to be illegal. At that point, the game is over. If your program is correct you win the game. If not, you lose the game.
• You must take a jump when required (we will follow the official rules).
• There is a maximum number of 100 moves per game, at which time a draw is declared. No other cycle detecting will be used.
• Programs will use a 7 piece endgame database so we can focus the neural network learning on middle game portion.