To assist you in deciding whether
to enroll for CS286, I've been looking through
possible projects for CS286 and there are several very promising
and practical ones. I have various papers on these ideas
and can generate copies for you to read if you would
like.
In general, the way I like to run the course is
by individual projects. There is only a weekly
course meeting if the students want it. Otherwise
we just have individual meetings, perhaps once
a week, or as required, some students like to check
with me every day, others like to meet every couple of
weeks.
Here is a list of project ideas. In each case the idea would
be to do a review of the psychological and neurological
literature, to design a model, implement it
and compare with existing data, and in each case
such a study would be publishable, so we would write
a paper at the end of the project. You
would learn this kind of programming, which is
not so hard, but I would also be involved
in the programming side. I view any
project as a collaboration with me leading to joint papers.
1. The Tower of London task. This is a variant of the
Tower of Hanoi problem and was designed for
diagonizing frontal patients. It's used in London
and the Institute of Neurology and studied there
also in at the Instititue of Cognitive Neuroscience,
and I'm in touch with all of these scientists.
There are psychological studies of how people
tend to solve the problem. Some MRI studies have
also been done. Our task could be to design
a brain model for the problem solving that people use.
2. The Tower of Hanoi problem. We did some work ast Caltech
on this last year, but there is a slot of scope for
more work notably in the learning that occurs
as people improve their strategies.
3. The Wisconsin card sort test. This is a very standard
neurological test used a lot for frontal patients.
A connectionist model has been published. Our task would
be to develop a brain model for how people try to solve this test.
4. Eye movement. We did get some data last year
on saccades as subjects solved Tower of Hanoi.
We have the system for doing further experiments.
It would be great just to devise a proper system
model for eye movement that takes into account
all the different levels of control and learning
that occurs. It would be nice to include
work from the Andersen lab.
5. Routine behavior. This is an interesting issue
for modeling of frontal or executive tasks. How
are actions made routine, how do we smoothly combine
routine with novel action. The neuroanatomy
involves the basal ganglia and there are tentative
theories of how routine action is learned
by them. This area is clinically important because
of parkinson's disease.
6. Action slips. There is by now some data on errors
made by people in execution of plans etc. For example,
pouring coffee into the sugar bowl instead of the cup,
or when getting dressed to go out for an evening
putting on your pyjamas and getting into bed.
Key authors are John Reason and Myrna Schwartz.
Explaining why people make the mistakes they
do, using a brain model, would be an interesting
project.
These are just some suggestions, you might have
some other ideas.
Also, if you would like to find out about previous
CS286 projects, there is a list on my websites,
and you could contact any of the students involved.
The student, David Devault, who did the eye tracking
research, was a senior and has just started at Rutgers,
but he and I are still in touch, and you can
reach him by his Caltech email devault@its.caltech.edu
The work of the summer students is also described
on my website and you could also talk with them, for
example Geoffrey Irving is at irving@its.caltech.edu