CS 517A - Machine Learning - Spring 2008

Text book:

• Tom M. Mitchell, Machine Learning, McGraw Hill, 1997

Location: McDonnell Hall 362

Time: Tuesdays and Thursdays, 2:30 pm - 4:00pm

Instructor office hours: Tuesday and Thursday after class, Jolley 506, or by appointment.

TAs and TA hours: Yunpeng Xu (yunpengtsu (at) gmail.com or xuyp (at) cse.wustl.edu), Tu 7-8pm, Jolley Hall 509

Brief description from the course catelog:  Formerly CS 527A. The field of machine learning is concerned with the question of how to construct computer programs that automatically improve with experience. Recently, many successful machine learning applications have been developed, ranging from data-mining programs that learn to detect fraudulent credit card transactions, to information-filtering systems that learn users' reading preferences, to autonomous vehicles that learn to drive. There have also been important advances in the theory and algorithms that form the foundation of this field. This course will provide a broad introduction to the field of machine learning.

How the course will be taught (which may affact how you take it):  I will not use slides for delivering lectures.  Instead, I will talk over and write on the board the main concepts, ideas, formulae and examples of the topics to be discussed.  This way the content can be delivered in a good pace so that students can follow relatively easily while taking notes and learn "on the fly".  This means that the students are not required to know the topics before the classes, but rather can learn directly from the class lectures.  This also implies that, although not required, the students are better off attending the classes.  To reinforce the knowledge learned in the classroom, the students will spend time on reading materials and homework assignments for each of the main topics of the course and finish a course project.  To help with learning, lecture notes will be available on line.

Schedule
Policies on homework, project, and grading
Collaboration Policy

Project description

Schedule  (Note: This schedule may be adjusted to meet the needs of the course.)

******** Introduction and background ********

******** Instance-based learning ********

******** Decision trees ******** 

******** Multi-layer perceptons (artificial neural networks) ******** 

******** Support vector machines ******** 

******** Evaluating hypotheses ******** 

******** Bayesian learning ********

******** Reinforcement learning ******** 

******** Combining multiple learners ******** 

******** Course projects ********

Policies on homework, project and grading

• Homework Assignments: There will be four sets of homework assignments, each of which covers two main topics of the total eight. The assignments are due either in the instructor's office in Jolley 506 by 2:15pm or in class in the beginning of the class on the given due date. Any homework submitted in class after 3pm will be given a 30% late penalty. If you arrive in class after 3pm, wait until the end of the class to bring up your homework. It is VERY disruptive to have someone walking in late and coming to the front of the class to submit homework. No assignments will be accepted after the instructor leaves the classroom. You are strongly advised to get started early so that you can get help if needed. These homeworks are not designed to be done in just a few days. There may be more than a single way to solve a given problem. You are expected to spend several hours for a two week homework assignment --- It will be very frustrating if you try to do this all in the last few days. For some problems you may need to think about them a while and then set them aside for a little.
• Project: There will be no exam for the course. Instead, we will have a course project for everyone. (The technical detail will be provided later.)  A project will have two parts:

• Reading some papers on a particular machine learning topic that we do not have time to cover or we do not go into detail in the class, and then giving an in-depth presentation to describe and discuss the problem, objectives, data used, procedure and techniques adopted, your criticism on the existing work, and your thoughts and suggestions on possible future research.

• Designing an algorithm/method for a particular datamining problem, most possibly from the paper(s) you read.

• Computation of the Final Grade: The following elements and scores will go into the final grade:
• Homework: 60 points total.
  
• Project:  40 points total.  10 for presentation, 20 for your design and implementation and 10 for report.

Policy on collaboration

Please keep any discussions you have with other students to a small group of no more than 3 students and be sure that each of you are equally involved. If you just listen in and are then able to understand and write up the solution you have missed at least half of the benefit of the homework. It is really important to work through the process of recognizing when you are heading the wrong way and learning how to work through the problem solving process.

Violations of any of the above rules will be dealt with harshly! The homework problems and projects are designed to help you learn the material being taught. Being told the solution and understanding it is VERY different from working through the process of actually finding a solution. If you do not take an active role in the process of solving the homework problems and project, then you won't get much out of it, hence you won't learn the material.