Final project for Artificial Intelligence, Fall 2015. Taught by Julian
Togelius at NYU Poly Tandon. Created by Whitney Mulhern, Giorgio Pizzorni, and Kunal Relia.
Part 1 of the project will involve implementing every single algorithm we learn in class.
Uninformed SearchBreadth FirstDepth FirstIterative Deepening
Informed SearchHill ClimberA*Simulated Annealing
Evolutionary AlgorithmsEvolution StrategyGenetic Algorithms
Adversarial SearchAlpha Beta Pruning
Supervised Learningk-Nearest NeighborPerceptron
- Decision Trees
- ID3
- Reinforcement Learning
- Q-learning
Part 2 of this project will involve setting off a Cambrian explosion in PacMan.
- Cambrianesque
- Life
- Yerking
- Darwinian
- Explosion
We will have six initially identical populations. They will all receive a different combination of (variance, temperature change). Variance will be either mutation, reproduction, or both. Temperature chance will either increase or be constant.
We will use simulated annealing to modify the randomness of our evolutionary algorithms. Half of the algorithms will become more random as time progresses, the other half will be left unmodified, as controls.
We will then perform some rudimentary statistical analysis on performance, but more interestingly, in how they evolve. How differently will they evolve? Will they evolve consistently through multiple trials? How will they evolve in response to different ghost controllers?