Learned about search problems (A*, CSP, minimax), reinforcement learning, bayes nets, hidden markov models, and machine learning.
Implemented Depth First Search, Breadth First Search, Uniform Cost Search, and A* Search. Created different heuristics. Helped pacman agent find shortest path to eat all dots.
Created basic reflex agent based on a variety of parameters. Improved agent to use minimax algorithm (with alpha-beta pruning). Implemented expectimax for random ghost agents. Improved evaluation function for pacman states.
Worked with Markov Decision Processes. Started with value iteration agent. Then, worked on changing noise and discount parameters to enact different policies. Built Q-Learning agent and an Epsilon Greedy agent. Then, used reinforcement learning to approximate Q-Values.
Implemented inference algorithms for Bayes Nets, specifically variable elimination and value-of-perfect-information computations, which reasons about the existence of invisible pellets and ghosts. The project includes constructing bayes nets, jointing factors, implemting variable elimination, normalization, marginal inference and value of perfect information.
Updated belief distribuition of ghost agents based on sequential noise readings and distribution of future ghost agent states. Moved pacman based on closest likely ghost location. Then, used particle filtering to achieve the same result. Finally, implemented joint particle filtering algorithm with multiple reactive ghosts.
Implemented a standalone perceptron. Built a mini-library for constructing and training neural networks. Built and trained several neural network to approximate sin(x), do digit classification, reinforcement learning and language identification.
