6.891 Machine Learning and Neural Networks


News

12/10 Last year's final... here.
10/29 Solutions to the midterm in pages: soln.ps.gz,soln.pdf.gz
10/7 The final will be on 12/13, at 1:30-4:30pm, in 3-270.

Summary

We will cover progress in machine learning and neural networks starting from perceptrons and continuing to recent work in "bayes nets" and "support vector machines". We will explore basic algorithms, including backpropagation, Boltzmann machines, mixtures of experts, and hidden Markov models. Throughout the relationship to statistical inference will be emphasized. Students will find that having had either 6.034 or 6.011 will be extremely helpful (i.e. familiarity with decision making and estimation in the presence of uncertainty and noise).

This is very likely to satisfy the AI graduate requirement for Area II graduate students in EECS.

This class is considered an elective in the Artificial Intelligence concentration required of EECS undergraduates.

Prerequisites
  • 6.041/6.042/18.313 (Probability)
  • One of the following is recommeded:
  • Information about Matlab can be found here. . Some of you may have taken 6.003 (which uses Matlab). They also maintain a web page describing MATLAB.

    Instructor
    Professor Paul A. Viola
    viola@ai.mit.edu (preferred point of contact)
    Room NE43-773 Phone x3-8828
    MIT AI Lab

    Teaching Assistant
    Kinh H. Tieu
    tieu@ai.mit.edu (preferred point of contact)
    Room NE43-771 Phone x3-7547
    MIT AI Lab
    Office Hours: Wednesday 3-4 PM and Friday 4-5 PM

    Lectures
    Time: Wednesday and Friday 11 AM-12:30 PM
    Location: 3-442
    Recitations
    There are two sections each week (of the same material, you need only attend one of them):
    Review Section
    Monday 3-4 PM (NE43-777)

    Format
    Class will consist of lectures and discussion of material coming from the textbook and related research papers. Discussion is strongly encouraged so it will help to have read ahead of lectures.

    Grading
  • There will be a mid-term and a final. Each will cover roughly one half of the course material.
  • Six problem sets will be assigned. Each will have exercises and programming assignments in Matlab.
  • There will be a final project. This is your chance to work on a machine learning project of your choosing. The scope of the project is roughly 2 problem sets.

  • Lecture Schedule
    Schedule: The schedule of lectures is not yet ready.
    Lectures Slides
    The lectures slides have been placed in this directory.
    Problem Sets
    The policy on self-grading is available...

    Pointers to related and clarifying information.
    Text
  • Duda, Hart and Stork, Pattern Classification
  • A collection of papers from the literature.