15-883 Final Project Instructions
Computational Models of Neural Systems
Each student taking the course for credit must complete a final
project. Students may design their own project in consultation with
the instructor, or do the assigned project described below. The
project will determine 20% of the course grade.
Student-Designed Projects
A student-designed project should explore some aspect of systems-level
neural modeling. The deliverables are: executable Matlab code, output
from sample runs of the code, and a writeup describing the model that
was implemented and the behavior that was observed.
Models do not have to be novel; it is acceptable to re-implement a
model described in the literature in order to better understand the
implementation and to do one's own experiments. It is also acceptable
to develop a novel model, or apply an existing model in a novel way.
Students must meet with the instructor to define the scope of their
project and the goals to be accomplished.
Assigned Project
The assigned project this year is to implement the Gluck and Myers
model of hippocampal-cerebellar interactions in classical
conditioning. This model uses both the simple Widrow-Hoff or LMS
learning rule and the more sophisticated backpropagation rule. You
have already seen working Matlab code for both these algorithms, so it
should not be difficult to adapt that code for your model.
Your model should reproduce the conditioning effects and lesion
effects discussed in the Gluck and Myers lecture: basic associative
learning, stimulus discrimination, sensory preconditioning, and
learned irrelevance, in both normal and hippocampal lesion animals.
Results should be displayed graphically as in Gluck and Myers chapter
6, i.e., bar plots comparing normal vs. lesioned animals, line plots
of the CRs, and line plots of the representational distance D between
stimulus vectors as training progresses. Follow the guidance in the
appendix to that chapter for constructing the simulation. Conduct
multiple runs in order to be able to calculate error bars for your
plots.
What to Hand In
Please submit your project as a zip file or a set of email
attachments. Your submission must include all of the following:
- Your MATLAB source code, so I can run your project if I want to test something.
- Plots and/or sample runs, as appropriate. The plots can be screen captures (JPEG or GIF)
or Postscript files.
- A writeup explaining what you did and the results that you
obtained. Did the model perform the way you expected? If you're
replicating someone else's model, point out any discrepancies
between your results and theirs. For example, how many trials were
required for training? (These differences might be due to differences
in minor parameters or implementation strategy. They won't harm your
grade as long as you are able to reproduce the qualitative character
of their results; you don't have to match them number-for-number.) Do
you have ideas about how the model could be extended or improved?
Due Date
The revised deadline for handing in final projects is Monday, December
17, at 5pm. But it would be better to get the project in sooner, to allow
adequate time for grading, or for feedback if some aspects of the
project need improvement.
Dave Touretzky
Last modified: Sat Dec 15 17:42:51 EST 2007