CSE 391 Special Topics in Computer Science -- Machine Learning
Spring 2012
Schedule
coming soon
Useful Links
Section Information
Section: 01
Date: Mondays and Wednesdays
Time: 5:20PM - 6:40PM
Room: S B UNION 237 WESTCAMPUS
Room: CS 2120
Instructor Information
Instructor: Luis Ortiz
Email: le [last name] at cs followed
by .stonybrook and finally .edu
Office: 2313-c
Phone: (631) 632-1805
Office Hours: TBD; for now, by appointment (please e-mail me in advance)
Course Description
A course on the fundamental concepts behind
intelligent systems that autonomously
learn to perform a task and improve with
experience. The course will cover learning frameworks and problem
formulations, standard models, methods, computational
tools, algorithms and modern
techniques, as well as methodologies to evaluate
learning ability and to
automatically select optimal models. Simple applications to areas such as
computer vision (e.g., character and digit recognition), natural-language processing (e.g.,
spam filtering) and robotics (e.g., navigating complex environments)
will motivate the coursework and material.
Overview
Roughly speaking, machine learning techniques strive to
automatically acquire expertise to effectively perform a task of
interest by efficiently processing task-related information, such as a
(usually large) data set of examples, and successfully extracting and
generalizing knowledge embedded within the available information.
There is growing demand for computer scientists with proficiency in
machine learning.
For example, the advent of technology for the
collection of vast amounts of digital data, such as that generated by
an ever expanding population of Internet users, has increased interest
in the development of machine learning software applications. Machine-learning-based technology such as driver assistance and
voice-activated systems in cars, automatic system personalization and
adaptation to individual user preferences and behavior, speech-driven
phone systems for customer service, speech-to-text capabilities,
recommender systems and e-mail spam filtering, are now commonplace.
Machine learning application areas include marketing, e-commerce,
software systems, networking, telecommunications,
banking, finance, economics, social science, computer vision, speech
recognition, natural-language processing, and robotics. Some problems
addressed using machine learning techniques include pattern
recognition and classification, knowledge discovery and data mining,
anomaly detection, credit/loan approval, credit-card fraud,
quantitative trading, automatic categorization of very large collections, such
as web pages, documents and images, effective ranking of web search
results (e.g., Google's PageRank), face recognition, tracking, machine
translation, and more recently, intelligent, adaptive control of virtual player behavior in
video games, smart debugging of computer programs
and memory management in operating systems. There is also recent interest in
creating computationally tractable machine learning tools for
recognizing and predicting general trends in individual or group
behavior in large populations, such as spending behavior, adoption of
new products, technology or habits, sharing in peer-to-peer systems,
and predicting the development of online communities within large
social networks such as Facebook.
Given the broad applicability of machine learning techniques, it is natural to expect the need for computer scientists with machine
learning expertise to continue to increase and expand in the years to
come.
This course covers the basic computational aspects of machine
learning, at an undergraduate level.
Course Objectives
This course provides an introduction to fundamental concepts and
modern techniques in machine learning, and prepare students for future
work in the area. The objective is to provide students with basic
knowledge and understanding of both the theory and practice of machine
learning, and to train students on the use and application of machine
learning ideas, paradigms and techniques.
Course Organization
The course format involves formal lectures, discussions and presentations, some
led by the students themselves. A list of recommended textbooks follows. Additional reading material
will be taken from a variety of sources, including other textbooks in machine learning and related
areas, tutorials and research literature in the area, as appropriate for an undergraduate level course.
Recommended Textbooks
- Stuart Russell and Peter Norvig. Artificial Intelligence: A
Modern Approach. Prentice Hall, second edition,
2003. (Machine-learning related chapters.)
- Thomas Mitchell. Machine Learning. McGraw Hill Higher Education, first edition, 1997.
- Richard O. Duda, Peter E. Hart and David G. Stork. Pattern Classification. Wiley-Interscience, second edition, 2001.
- Christopher M. Bishop. Pattern Recognition and Machine Learning. Springer, first edition, 2007.
Course Project
Students complete a project on an
application of machine learning to a particular problem, which the students
choose in consultation with the instructor. The chosen course project
requires instructor's approval. The project should have an experimental component. Ideally, the project will address a new problem and produce a
novel application. Students make an oral presentation of their project
proposal. To monitor the project's development, students periodically
make oral presentations as the project progresses. Students produce
(and hand in) a final written report on their project and give a
final project presentation by the end of the course term.
Student Evaluations (Tentative)
Students are evaluated on their performance on homework assignments
(25%), a midterm (15%) and a final exam (20%), their participation in class discussions (5%) and the quality of their project and respective presentations (35%).
Tentative List of Topics
- Machine-learning fundamentals: classification, regression
and clustering; noisy, noise-free and incomplete data; supervised
and unsupervised learning; learning under uncertainty; hypothesis classes, model complexity, model selection, Ockham's razor and the bias-variance dilemma; generative vs. discriminative probabilistic models; dynamic environments, reinforcement learning and the exploration-exploitation dilemma
- Basic models and methods: nearest neighbors, decision trees, linear discrimination, neural networks, support vector machines (SVMs), boosting and bagging, naive Bayes classifiers, gradient-descent, Q-learning
- Advanced topics: expectation-maximization (EM), Hidden Markov
Models (HMMs), K-means clustering, mixture-of-Gaussians, component analysis
NOTE: The list of topics, as well as the emphasis
on each topic, may vary slightly depending on the background and interests of the course participants.
Schedule
coming soon!
Useful Links