This is an advanced undergraduate course that covers advanced topics in Artificial Intelligence. Topics vary from one offering to the next and are likely to be drawn from the following list:  planning, scheduling, games, search, reasoning (constraint-based, model-based, spatial, temporal), knowledge representation, decision-making under uncertainty, reinforcement learning, agents, foundations.

On completing this course students will have a deep understanding of the taught topic(s):

Planning is concerned with finding a sequence of actions that achieve a predefined goal.  Students will learn key planning algorithms as well as the main applications of planning.

Games: Game playing algorithms have been developed for most important games. The quality of their play varies quite a lot though. While the world chess master is a computer, Go programs have no chance even against mediocre players. The student will learn state-of-the-art techniques for selected games.

Search: The solution of many AI problems, in particular planning, scheduling, and game problems, is based on extensive search in solution space. Students will learn sophisticated AI search algorithms, in particularly exact and heuristic methods for pruning the search space.

Scheduling is concerned minimizing the time required to complete a set of interrelated jobs, each of which consisting of a sequence of actions, where each action has a given duration and might require some resources. The student will learn representative algorithms and apply them to representative problems like time-tabling.

Spatial and Temporal Reasoning is concerned with different ways of representing spatial and temporal knowledge and with techniques for reasoning about this knowledge and tools and methods for analysing reasoning algorithms.

Knowledge representation: Knowledge-based systems have to internally represent knowledge about the world. The choice of representation is influenced by the required expressivity and the required efficiency of doing inference. The student will learn how to represent knowledge for complex real-world problems.

Decision-making under uncertainty is concerned with planning for noisy or uncertain problems.

Reinforcement learning is concerned with how an agent ought to take actions in an environment so as to maximize its long-term reward. It deals with the problem of sequential decision-making in unknown environments. Classical applications are algorithms that learn to play games like checkers, backgammon, and recently Go purely by self-play.

Agents: Most AI problems can be phrased as one or multiple agents interacting with each other by sending and receiving signals. The student will learn how to classify and design multi-agent systems and study their dynamics.

Foundations:The grand goal of AI is concerned with building systems exhibiting generic intelligence, rather than solving specific problems like chess. This raises the questions of what intelligence is and whether machines can ever surpass human intelligence. Students will learn the necessary philosophical, statistical, and computational concepts that allow to address these difficult questions.

Assignments (45%); Seminar (15%); Final Exam (40%)

Thirty one hour lectures

COMP3620

Parts of

Marcus Hutter (2005) Universal Artificial Intelligence, EATCS, Springer.
http://www.springer.com/computer/ai/book/978-3-540-22139-5

Shane Legg (2008) Machine Super Intelligence, Lulu, PhD thesis
http://www.lulu.com/content/2043514

http://cs.anu.edu.au/courses/COMP8620/