ECE/CS 547: Neural Networks

sciencediscussionΤεχνίτη Νοημοσύνη και Ρομποτική

20 Οκτ 2013 (πριν από 3 χρόνια και 9 μήνες)

77 εμφανίσεις

1

Fall Semester 2003


ECE/CS 547: Neural Networks


Instructor:


Prof. Thomas P. Caudell


Office:
ECE Rm 235D


email: tpc@e
ce.unm.edu


Phone: 277
-
5637


Goals:

The primary goals of this class are as follows:

a)
To develop an understanding of the underlying engineering principles, design, and
operation of biological nervous systems as a basis for the development of artificial
neural networks.

b) To develop the theory and analysis of a variety of neural models,
systems, and
learning environments.

c) To develop a series of software simulations of neural networks to aid in the
understanding of their functionality.

d) To prepare students to apply neural networks to real world applications through
researching and
writing a conference level paper.


Description:

The operational principles found in the nervous systems of animals will be
used to motivate the design of artificial neural networks. These artificial models will be
analyzed and their theory developed. Top
ics will include an introduction to
neurobiology, simple models, learning processes, the Perceptron, LMS Algorithm, the
Multilayered Perceptron, radial bases nets, self organization, neuro
-
dynamics, adaptive
resonance theory, electronic and optical hardwar
e implementations, operational systems,
and applications.


Prerequisite:

First year engineering, computer sciences, math, psychology, or basic
sciences grad student standing. Undergraduate students are welcome by permission of
their home departments and t
he instructor. Calculus, differential equations (first order),
linear algebra, geometry, basic statistics, computer programming skills in C/C++.


EMAIL:

You must have an email address. Homework assignments and other class
related communications will be
done electronically.


Office Hours:

I encourage you to come by during these hours to discuss neural network
related topics or to ask questions about assignments. Other times are available by
appointment.

Please schedule times with class graders for assistance with programming
assignments.


Primary Textbook:

(Required)

"Neural Networks:

A Comprehensive Foundation", by Simon Haykin, Prentice Hall.
(1999). Second Edition. ISBN 0
-
13
-
273350
-
1


Topics:

2

-

History of the field

-

Introduction to biological neural networks as motivation for this form of computation

-

Survey of learning processes

-

Perceptron

-

Multilayer Perceptrons

-

Radial Basis Functions

-

Support Vector Machines

-

Committee Machines and Boosting

-

Self Organizing Systems

-

Neurodynamics


Other Material:

(Optional)

-

"Principles of Neural Science", Eric Kandel, James Schwartz,

and Thomas Jessel,
Elsevier Science Publishing, New York (1991).

-

"Vehicles: Experiments on Synthetic Psychology", Valentino Braitenberg, MIT
Press (1987).

-

"The Computational Brain", Patricia Churchland and Terrence Sejnowski, MIT
Press (1993).

-

"Anal
og VLSI and Neural Systems", Carver Mead, Addison
-
Wesley (1989).

-

Selected research papers.


Guest Speakers:

As opportunity presents itself, guest speakers may be invited to lecture on special topics.


Grading:


Homework (book and simulation)

6
0%


Researc
h Paper

4
0%


Research Paper:

Eight or less pages, typed, 12 pt Times font, spaced single and a half,
single sided, single column, 1” borders around, including references, fi
gures, and tables.
Eac
h student must inform the instructor

the
topic of their paper as early in the semester as
possible. Paper must deal with a research topic in neural networks, which may be based
on library research, analysis and/or a programming project. The following are example
topic areas:


-

analysis of an
application


-

literature searches and reviews


-

model selection


-

preprocessing of data


-

simulations


-

software/hardware issues


-

user interfaces


-

system integration


-

evaluation of performance


-

other

Write this paper as if you were going to
submit it for publication to a conference. Paper
format example and more detailed guidelines will be provided later in the semester.


3

Homework:

Throughout the semester, assignments will be given from problems in book
and to program and experiment with spe
cific but simple neural algorithms. All of the
homework and test data sets will be distributed to you over email. An
optional
open
source simulation tool will be provided.


Class Participation:

Please participate in class by asking questions, making relevant
comm
ents, and reading the material
before

class.