Announce
CS491/591-005: Introduction to Machine Learning
Fall, 2002
Tue/Thu 4:00-5:15
ME 218
Instructor: Terran Lane
Course Announcement
How can computer programs learn from experience? Can we teach robots to behave rather than explicitly program them? Or build systems that learn to recognize faces from images rather than hand-tune individual face detectors? What about the discovery of deeper relations -- can we detect the presence of "hidden variables" and causal structures in biological data, for example? How long does it take to get good at a performance task? What are the tradeoffs between representational power and learnability? Are there limits to what can be learned? And what kinds of fun things can you do with adaptive learning systems once you have them?
These are some of the questions addressed by the field of machine learning. In the past two decades, systems employing ML have demonstrated performance often rivaling or excelling that of humans or hand-programmed systems in applications as diverse as protein modeling, astronomical observation classification, speech recognition, medical diagnosis, microprocessor branch prediction, visual object recognition, game playing, robotic navigation and control, web page analysis, computer security, handwriting recognition, fraud detection, image enhancement, and stock market analysis.
In this course, we examine the principles and practices of algorithms that improve their performance through experience. Topics include fundamentals of representation; the statistical learning model and PAC learning; performance evaluation; classification and supervised learning; generative versus discriminative models; unsupervised learning and the expectation-maximization procedure; time series analysis; and reinforcement learning. Discussion will be augmented with a number of application examples and readings from the current literature. Students will have the opportunity to develop programs with learning components and to follow up topics of personal interest in a reading or development project.
Pre/co-requisites: Knowledge of algorithms and complexity analysis (CS 461 or equivalent) and probability/statistics or by permission of the instructor. Knowledge of artificial intelligence and/or pattern recognition helpful but not essential. Available for graduate or undergraduate credit.
