EEE511: Artificial Neural Computation Systems

Last updated 12-09-2007
Press shift-reload to see the latest version.

Logistics

  • Instructor: Kari Torkkola
    • Contact: torkkola [at] asu [dot] edu
    • Office: Any location serving decent coffee
    • Availability: By appointment (send me email), sometimes after lectures
  • Next class
    • Location: BAC328
    • Time: MW 9:15-10:30
    • Begins: August 20, 2007
  • Grading:
    • 5 homework assignments, 70% total (late homework is not accepted)
    • Final project 30%
    • Filling out this questionnaire as soon as possible would be extremely helpful. It will aid me direct the rest of this class towards your interests and needs.
  • Delivery of handouts, homework, and other material:
    • All material will be delivered via the class web page http://www.eas.asu.edu/~eee511 .
    • If there is something you are required to have for the class, it will be downloadable before 9 pm the previous evening. Lecture notes are not guaranteed to be downloadable until about the midnight before the class.

Contents

This course covers the principles of artificial neural networks (ANN), collective computational phenomena emerging from simple interconnected elements.  The emphasis is on word "artificial" (as opposed to real); this class will not reveal to you how the brain works.

Rather than to concentrate on the details of each ANN model (and there is a fair number of them), the purpose is to obtain a coherent overview of the field, an ability to find and digest deeper knowledge in any particular subfield, and most importantly, ability to apply an appropriate type of an ANN model to a given problem. However, this does not mean that we will be sloppy with the basics! We will attempt to maintain a firm grounding in statistics.

There will be five sets of homework. Each one will involve problems, some programming and experimentation.  The course will culminate in a final project - a largish real problem, preferably of your choice and from a domain close to your heart. The instructor is happy to give out topics close to his heart.

Prerequisites

  • differential multivariate calculus (gradient, chain rule, ...)
  • linear algebra (eigenvalues, singular value decomposition, ...)
  • probability and statistics (probability densities, Bayes rule, ...)
  • ability to program in MATLAB

All of the above can be compensated for by interest and effort, and by reading these primers:

Announcements

  • Grading including the project paper  is posted

Course Material

 

Software

ANN simulation software

Matlab help

 

Demos

  • Here is a pointer to the work of Karl Sims about evolving artificial creatures. You can find the movie shown at class after following a couple of links from there, or a local copy here. Two papers of his describing the creatures can be found from his page, too.
  • Dynamical & Evolutionary Machine Organization at Brandeis University contains " Exact Representations from Feed-Forward Neural Networks " that will be shown at class. Take a look also at other interesting stuff in the pages.
  • LeNet5, convolutional networks for character recognition by Yann LeCun.
  • SOM demo with colorspace input. Local copy here, and original website here.
  • Two demonstrations of solving a combinatorial optimization problem using one-dimensional neuron arrangements. Travelling salesman problem solved using Self-Organizing Maps, and same problem solved using SOM-like Elastic Net. Both pointers are local copies.
  • A demo about Growing Self-Organizing Networks from the Ruhr-Universität Bochum in Germany.
  • Support Vector Machine demonstrations at Royal Holloway University of London.

Useful Resources

Journals

These usually allow searching old issues.


Conferences


 

Finding relevant papers in the web


Data Sets