COMP 424 - Artificial Intelligence

• An artificial intelligence (AI) course that I took during my fifth year at McGill. Topics included:
  • Uninformed Search - state space graph, search tree, BFS, DFS, uniform-cost search, depth-limited search, iterative deepening search
  • Informed Search - heuristics, best-first search, heuristic search, admissible heuristics, consistent heuristics, A* search, dominance
  • Local Search for Optimization - hill climbing, simulated annealing, parallel search, local beam search, genetic algorithms
  • Constraint Satisfaction Problems - constructive vs local search approaches, backtracking search, constraint graph, arc consistency, AC-3, forward checking
  • Searching Under Uncertainty - partial observability, unobservability, non-deterministic actions, conformant planning, AND-OR search, belief states
  • Game Playing - minimax search, evaluation functions, alpha-beta pruning, forward pruning, monte-carlo tree search (MCTS), upper confidence trees, exploration vs exploitation tradeoff
  • Knowledge Representation and Logic - propositional logic, propositional inference, model checking, normal forms (CNF, DNF, Horn), validity and satisfiability, forward chaining, backward chaining, first-order logic (FOL), quantifiers, FOL inference, FOL proofs, resolution, skolemization
  • Planning - STRIPS planning, progression/forward planning, regression/backward planning, planning as logic
  • Modelling Uncertainty - bayesian probability, probabilistic inference, naive bayes model, bayesian networks, inference in bayesian networks, variable elimination algorithm, bayes ball algorithm
  • Learning in Bayesian Networks - maximum likelihood estimation (MLE), laplace smoothing, latent variables, learning with missing data, expectation maximization (EM), soft vs hard EM, k-means clustering
  • Causal Graphical Models - causation vs correlation, do-operator, causal vs probabilistic conditioning, identifying causal effects, back-door paths
  • Utility and Decision Theory - axioms, lotteries, maximum expected utility (MEU), decision graphs, value of information
  • Bandits - bandit problems, greedy bandits, epsilon-greedy bandits, softmax action selection, optimistic initialization, upper-confidence bounds, contextual bandits
  • Markov Decision Processes (MDP) - sequential decision-making, MDP framework, policy iteration, policy evaluation, bellman equation, value iteration, reinforcement learning
• For the final project, we built an agent for the board game Colosseum Survival. The project was done in pairs and my partner and I implemented ours using Monte-Carlo Tree Search equipped with some domain-specific heuristics. The project code and report can be found here and a demo of our agent beating a random agent is shown below.