CS-3400/CogS-4680: Neural Nets and Intelligent Machines

Spring, 2022

Instructor: Dr. Tom Carter
DBH-287       667-3175
tom@cs.csustan.edu or tcarter@csustan.edu
https://csustan.csustan.edu/~tom

Texts: There is one required text and one suggested text for this course:

     Required: Deep Learning with Python
          by Francois Chollet
          (ISBN-13: 978-1617294433)

     Suggested: Computational Neuroscience - a First Course
          by Hanspeter A. Mallot
          (ISBN-13: 978-3-319-00860-8)

We'll talk about other resources for the course during the semester in class . . .

I will be using a "hybrid" approach for this class. Some work will be through the campus Canvas system, but I also plan to do in person classes (when possible) and "synchronous" Zoom sessions, where I can present some material, and we can discuss relevant topics. Our first Zoom session will be Tuesday, February 2, at 9:30 am. I will provide more info about the Zoom session soon.

Objectives:

A long-standing hope has been that computers will develop into intelligent assistants for us. Traditional approaches to computing (under the general description of von Neumann machines) have not led to particularly successful implementations of ``intelligent behavior.'' In recent years, several new approaches to computing have shown strong potential. In general, these approaches attempt to learn from the biological world which has led to human intelligence, and to apply abstractions from the biological world to the artificial world of computers.

Our primary focus will be on neural network approaches to computing, but we will also spend some time on other topics such as genetic algorithms and ``artificial life.'' The lab component of the course will involve hands on work with existing implementations of the various approaches to biologically inspired computing, or development of a new project, or research into other possible approaches.

The texts take a relatively mathematical approach to the topic, so be prepared -- but as usual, I will work to make the material comprehensible to all of us ...

Topics:

• Biological neurons and neural structures
• Classical models of neural functioning, such as Hodgkins/Huxley and perceptrons
• Multi-layer feed-forward nets with back propagation of errors
• Unsupervised learning, such as Kohonen networks
• Elements of machine learning
• Contemporary approaches such as deep learning and convolutional models
• Alternative biologically inspired approaches such as genetic algorithms and "artificial life"
• Social, philosophical, and other issues and concerns related to the development of "artificial intelligence"

Grading:

The grades for this course will be based on three components: written responses to readings and brief exercises, a midterm exercise, and project(s). My expectation is that some of the projects will be developed by teams.

Each week, I want you to submit on Canvas (look in the "Assignments" section) a brief response to the readings for the week. We'll discuss this more as we move forward in the semester . . .

The components will be weighted approximately equally.


The work you do for this course will be your own. You are not to submit other people's work and represent it as your own. However, I do expect and encourage you to work collaboratively with others during the course.