Final Project

It's time to begin thinking about your final project! You can choose to do essentially anything inside the area of AI, so long as I approve it in advance. Submit a proposal of no longer than one page telling me what you want to do. You should have enough detail to convince me that you've given it some thought. You can build on material we do in class, or you can use it as an opportunity to learn about subject matter we don't have time to cover. Make sure you check out the course syllabus to see how much the project is worth: it may be a smaller portion of your grade than you expect. It is my expectation that you should put about as much work into this as you have for one of our large assignments, and so I proportioned the grade accordingly.

Make sure that when you submit your final project, you also submit an explanation of what you have done, how I should run your code, and how I should interpret your results. This does not need to be long. It should merely prevent a problem I've had in the past: students have submitted work that either I can't run, or it runs but I can't figure out what it's doing.

Here is a list of some projects that people have done for for this class in the past. In the past, when I haven't provided this list, students have begged me for suggestions. In other past years, when I have provided this list, students agonized and felt compelled to come up with a project that wasn't on it. That was their choice, not mine -- it is TOTALLY WONDERFUL AND NOT BAD if you decide to do one of the projects on this list! Keep in mind that I vary the topics a bit each year and the order that I cover them, so we may not have covered some topics yet in this list. Similarly, there are topics we may have covered that you may not find in this list. Some of these sound harder than they are; others seem deceptively simple, but require significant amounts of work at the end to actually make them work properly.

• Implementing Knn to Classify Bedform Stability Fields
• Blackjack Using Genetic Algorithms
• Computer game players: Go, Poker, Scrabble, Magic: The Gathering, Mastermind, Cribbage (be careful, shouldn't be a rehash of Othello)
• Reinforcement learning in playing Bridge
• Computer puzzle solvers: Minesweeper, mazes
• Pac-Man with intelligent monsters
• Automated 20-questions player
• Paper on planning
• Neural network spam filter
• Learning neural networks via genetic algorithms
• Solving neural networks via backprop
• Code decryptor using genetic algorithms
• Box pushing agent (competing against an opponent)
• Evolving a perfect candidate
• Motion planning via A*
• Evolving optimal auction strategies
• Evolving Othello strategies

Here's a warning: Genetic algorithms are a popular choice for final projects, and they work great. But they often take a significant amount of time to run: you need to leave many days to play with the crossover/mutation/etc operators, the size of the population, and so on. That's not intensive time for you personally, but you need to have the time on the clock to let it run and occasionally futz with it. GA projects can be disasters as "last minute projects," i.e. if you don't plan for many days of letting it run and occasionally tweaking it.

Here's a list of topics that didn't work so well:

• Too complicated games: Risk, Yahtzee, Scrabble, Chess, Battle Simulation
• Got too focused in making game work
• I sometimes had trouble running the game
• Game was often incomplete
• Didn't have time to do enough AI
• Problems that were too vague
• Simulated ant colonies / genetic algorithms
• Bugs swarming for heat (emergent intelligence never happened)
• Emergent ants via game of life
• Finding paths through snow
• Boosting on protein folding data
• Couldn't get boosting working right, needed more time on small datasets (spent lots of time parsing protein data)