Final Examination |
CSE 415: Introduction to Artificial Intelligence The University of Washington, Seattle, Winter 2023 |
Date: Tuesday, March 14 (2:30-4:20 PM) |
Format: The format of the final exam will be similar to that of the midterm exam. However, the exam will be longer. The topics covered will be drawn from the following, which includes some topics from the first part of the course and some from the second. |
Topics:
Turing Tests State-space search States, operators, combinatorial explosion Admissible heuristics, Consistent heuristics Domination of one heuristic over another A* search Minimax search for 2-player, zero-sum games Static evaluation functions Alpha-beta pruning Expectimax search Markov Decision Processes States, actions, transition model, reward function Values, Q-states, and Q-values Bellman updates Policies, policy extraction, number of possible policies Value Iteration Reinforcement Learning Q-learning Application to the Towers-of-Hanoi puzzle and Grid Worlds Probabilistic reasoning Random variables The joint probability distribution Marginal probabilities Conditional probability Product rule, chain rule Independence of random variables COnditional independence Bayes' rule Bayes nets Representation using graphs, marginals and conditional distributions Number of free parameters Inference in Bayes nets D-Separation in Bayes nets Perceptrons How to compute AND, OR, and NOT Simple pattern recognition (e.g., 5 x 5 binary image inputs for optical character recognition) Training sets, training sequences, and the perceptron training algorithm Linear separability and the perceptron training theorem Multi-layer perceptrons with soft thresholds (e.g., standard logistic function) Autoencoders Markov Models Transition probabilities Belief vectors The Mini Forward algorithm Stationary distribution Hidden Markov Model structure Using Bayes Rule to update belief when an observation is made Natural Language Processing Probabilistic Context-Free Grammars Scoring parses with negative log probabilities Representing documents using vectors based on bag-of-words modeling The Future of AI Asimov's three laws of robotics, Kurzweil's "singularity" The Trolley Problem, Empathy as a goal for ethical AI. Bias in ML -- esp. facial recognition Pros and cons of advanced artificial intelligence Superintelligence and Bostrom's concept of singleton |