Lecture: 3 hours

Project: 1 hour

Prerequisite: Course ECS170

Grading: Letter; paper reviews (20%), project (50%), final (30%)

Catalog Description:
Artificial intelligence techniques for knowledge acquisition by computers. Fundamental problems in machine learning and discovery. Systems that learn from examples, analogies, and solved problems. Systems that discover numerical laws and qualitative relationships. Projects centering on implementation and evaluation.

Machine learning is concerned with the question of how to construct computer programs that automatically improve with experience. The goal of this course is to present the key algorithms and theory that form the core of machine learning. The course tries to strike a balance between theory and practice.

Expanded Course Description:

  1. Overview; claims; ways to evaluate learning and discovery systems
  2. Inductive learning (learning conjunctive concepts from examples)
  3. Learning decision trees
  4. Conceptual clustering
  5. Learning and discovery of heuristics for problem-solving
  6. Discovery of numerical laws
  7. Learning from analogies, learning from experiments
  8. Deductive learning (explanation-based learning and the connection to program optimization)

T.M. Mitchell, Machine Learning, McGrawHill, 1997

Each student is expected to do a project using one of the machine learning algorithms described in the text book and applying the method to solve a realistic problem. The student is strongly advised to pick a problem domain that is very familiar to him/her and has knowledge about where to get the data. A student’s proposed thesis or dissertation topic is the best area to choose such a subject area and ask the question, “what if I solve this problem using machine learning?” As each student is free to choose his/her own machine Learning method before the classes begin, it is important that the student has some idea as to the method(s) under consideration.

The student is then expected to write a project report by following the same general guidelines one follows in writing a conference paper. Typically, this paper shall be between 6-10 pages long (not counting any program listings the student may wish to submit as a backup) and should contain all information necessary to understand the paper and replicate the results – if one so wishes.

Instructors: K. Levitt, R. Vemuri

Prepared by: R. Vemuri (September 2002)

Overlap Statement:
This course does not have a significant overlap with any other course. It covers some of the topics as in EEC 207, but does so at a more software-related level. Applications in computational science are emphasized throughout.