Connectionists: Summary of courses available

[Geoffrey Goodhill]

Dear Connectionists,

Thank you very much to everyone who replied to my request for
information about courses being taught in Computational
Neuroscience. Below is a lightly-edited summary of responses so far
(sorry for the delay), in roughly the order they arrived. I have also
made this file available at

www.goodhill.org/comp_neuro_courses

I will be happy to update this file periodically if you would like to
send me additional info.

Thanks,

Geoff

-------------------------------------------

BEDNAR (Edinburgh)

"Computational Neuroscience of Vision"

http://www.inf.ed.ac.uk/teaching/courses/cnv/

Text: Miikkulainen et al, Computational Maps in the Visual Cortex,
Springer (2005).

Officially, the course was at the MSc level, although there were a
couple of 4th-year undergraduates also. There are 18 1-hour
lectures.  This year (the first year it has been offered) there were
17 students who took it for credit, plus three auditing.

There are no specific prerequisites, because it's not particularly
mathematical, and because I include a long intro to biological vision.

Assessment: 2 practical assignments, plus a short-essay exam.

"This course focuses on understanding the computational mechanisms
underlying animal visual systems that are similar to those of
humans. The main emphasis is on how the properties of neurons across
the two-dimensional surface of the visual cortex are organised
topographically to represent and transform the relevant features of
visual stimuli. Because the visual cortex is the primary model system
for understanding the cortex in general, the course also acts as an
introduction to computational processing in all topographically
organised cortical regions."

ROSSUM (Edinburgh)

Graduate
http://www.inf.ed.ac.uk/teaching/courses/2005/nc.html
Text: Dayan & Abbott, Theoretical Neuroscience, MIT Press, 2001.

"This module aims to examine: How the brain computes and processes
information from the outside world.  How the brain wires up and how it
stores information.  We will study the brain at a fairly low level, so
that we can make contact with neurophysiological data. We will show
the necessary biological data and how it can be described in
mathematical terms. We will present modelling methods applicable to
various levels of organisation of the nervous system (e.g. single
cells, networks of cells). We discuss models of particular brain
subsystems.  In the practical session we use Matlab and NEURON to
simulate the models (No familiarity with NEURON required, some self
study of Matlab is beneficial.)"

BLACKWELL (GMU)

http://www.gmu.edu/departments/krasnow/CENlab/syllabus.html

Text: Johnston & Wu, Foundations of Cellular Neurophysiology, MIT Press,
1995.

"An intense review of neurobiology for graduate students interested in
studying how nerve cells integrate and transmit signals, and how
behavior emerges from the integrated actions of populations or
circuits of nerve cells. The course covers electrical and biochemical
properties of single neurons, and electrical and chemical
communication between neurons. Emphasis is on mathematical
descriptions and computational techniques used to study and understand
neurons and networks of neurons.Learn how to ask and answer questions
about neurobiology using computational techniques"

http://www.gmu.edu/departments/krasnow/CENlab/csi735.html
Text: Dayan & Abbott, Theoretical Neuroscience, MIT Press, 2001.

"An intensive introduction to systems neuroscience. The anatomy,
physiology, and function of each of the major brain structures and
systems will be presented. The emphasis will be on behaviors that
emerges from integrated actions of populations of
neurons. Computational techniques used to study and understand
networks of neurons also will be addressed. Students are expected to
do assigned readings prior to class and to participate in class
discussions (20% of grade) and present a project in class (80% of
grade)."

TRAPPENBERG (Dalhousie)

Graduate
Text: Trappenberg, Fundamentals of Computational Neuroscience, OUP, 2002.

The role of modelling in neuroscience.
Example of mapping networks (e.g. simple perceptron or MLP) to recognize
letters.
Neurophysiological basis of spiking neurons.
Models of spiking neurons.
Modelling population averages.
Synaptic Plasticity (spike time and rate models)
Overview of some organizations in the brain (e.g. very basic anatomy,
cortical maps, laminar organization, etc).
Memory and learning.
Recurrent attractor models (typically point attractor networks, sometimes
continuous attractors as an example of my research specialty), Hippocampus
Self-organizing maps, typically in the form of Willshaw and Von der
Marlsburg).
Some basic reinforcement learning.
Usually at least one example of a coupled network or system level model of
the brain.

The course also includes some technical introductions, in particular
Introduction to Matlab
Numerical integration of differential equations
Basic vector and matrix algebra
Sometimes some statistics

No perquisites are required and the course is introductory in nature due to
the diverse background of the students, but I make it clear that they need
to write basic simulation programs that typically involve the numerical
integration of ODEs.

A major part of the course and the evaluation are individual student
projects.

Finally, we will offer a 4th-year undergraduate elective in the UG
neuroscience program for the first time this fall. This course will be along
similar lines as described above (and initially co-taught) but with more
specific projects instead of the very research-oriented projects required by
the grad students. I also taught a weekly section of the main graduate
neuroscience course on the topic of computation neuroscience but did not
continue this in the last two years due to my schedule restrictions.


BOWMAN (Kent)

Graduate/undergraduate
http://www.cs.kent.ac.uk/teaching/05/modules/CO/6/36/index.html
Texts: Reilly & Munakata, "Computational Explorations in Cognitive
Neuroscience, Understanding the Mind by Simulating the Brain", MIT Press,
2000.

Introduction to cognitive neural networks
The individual neuron
Networks of Neurons
Model Learning
Task Learning
Combined model, task learning and other mechanisms
The brain and implications for biologically plausible neural networks
Perception, Vision, Object Recognition and Attention


ERMENTROUT (Pittsburgh)

Here are 2 classes I've taught several times over the years.

I currently use Abbott & Dayan for the comp neuro course and I use a draft
of my book, Math Neuroscience, for the math course.

http://www.math.pitt.edu/~bard/bardware/classes/compneuro/neurointro.html
http://www.math.pitt.edu/~bard/bardware/classes/introcns.html

http://www.math.pitt.edu/~bard/bardware/classes/mathneuro/intro.html
http://www.math.pitt.edu/~bard/bardware/classes/mathneuro/mn.html
http://www.math.pitt.edu/~bard/bardware/classes/mathneuro/mn06.html


FELDMAN (ICSI)

http://www-inst.eecs.berkeley.edu/~cs182/sp06/

UG class, 15 weeks, ~70 students

Text: Feldman, From Molecule to Metaphor: A Neural Theory of Language,
MIT Press, 2006.


HINTON (Toronto)

Undergraduate
http://www.cs.toronto.edu/~hinton/csc321/index.html

Graduate
http://www.cs.toronto.edu/~hinton/csc2535/index.html


SANCHEZ (Florida)

I teach a course titled "Fundamentals of Computational Neuroscience" every
spring semester at the University of Florida. This past year, I had 8
graduate students. You can see the course website here

http://nrg.mbi.ufl.edu (select courses)


DAYAN (UCL)

Graduate, 24 lectures, 5-10 students.
Text: Dayan & Abbott, Theoretical Neuroscience, MIT Press, 2001.

Prerequisites: facility with standard mathematical methods for
physcists, plus some probability theory.

neural encoding and decoding, information theory
neuroelectronics (spiking, dendrites, propagation,  synapses)
development and learning

Assessment: brutal [sic]

The main problem is deciding how to teach systems as well as
methods. we currently solve that by having a separate course.


REGGIA (U Maryland)

I teach the following course roughly every two or three years in the
Comp. Sci. Department at the University of Maryland:

   Title: Neural Computation
   Level: graduate students
   Duration: about 30 lectures
   Size: about 20 - 25 students per offering
   Prerequisite: graduate standing in comp sci/EE/neuro/psych/etc.
   Topics: methods/algorithms for neural computation, with a heavy
      emphasis on learning, including perceptrons, backpropagation,
      radial basis function networks, associative nets, unsupervised
      learning, self-organizing maps, associative memories,
      dynamical system theory, oscillatory nets, etc.
   Textbooks: several as references plus papers from the literature
   Assessment: typically four programming projects, homework problems,
      midterm and final exams


GRAHAM (Stirling)

I teach an undergraduate computing science honours-level
half module called "Computing and the Brain" (point your browser at:
www.cs.stir.ac.uk/courses/CSC9YF/

This normally has 10-20 students, uses the NEURON simulator
and loosely refers to Churchland & Sejnowski, "The Computational Brain"
and Lytton, "From Computer to Brain".
It assumes nothing (just as well!) from the students and provides
a basic intro to neuroscience/computational neuroscience.


SWINDALE (UBC)

I teach an introductory course in computational neuroscience aimed at
graduate students (it is a 500 level course), it is for 3 credits and
is one session of 3 hours per week for one term. There are typically 5
- 10 students in the class and typically several of them are auditors
who contribute in exactly the same way as those doing it for credit.

Prerequisites: Some background in either physics/math/computer
science/engineering and/or some neuroscience background. This means
the backgrounds of the students are usually quite mixed. More often
than not they lack the neuroscience side, so the course usually
includes about 3 - 4 classes of introductory neuroscience material, of
course stressing the quantitative aspects.

In the first class we ask 'is the brain a computer?' (Turing's paper)
and 'what is computation?', for which we go to Penrose for a good
explanation of Turing machines and computability. That usually gets
the students talking a lot.

After that we begin for real with nervous system basics including
cellular biophysics, synapses, receptors, compartmental modelling,
very detailed 'Neuron' type simulations. Then we go on to look at
simplified model neurons and some basic types of neural net, including
various associators, adalines, Hopfield, back-propagation etc; then we
usually look at some more specifically neural models e.g. cortical
maps (correlation based learning, Kohonen, maybe even elastic net);
after that the selection is somewhat up to the students, this year we
had a class on sparse coding and usually we review something to do
with hippocampus. In the last class we return to the beginning in a
sense and look at the neural corelates of consciousness.

Textbook: There isn't an ideal one. Dayan & Abbott is a great book but
is a bit too advanced to be used for the whole course (I have started
to use parts of it though). Mostly I use chapters from various
sources, Anderson is the best book for the basic simple network models
that we look at. However it does not cover the neuroscience well
enough for me, also I find the mathematical notation is
over-simplified to the point that it makes it harder (for me at least)
e.g. the section of back-prop which is tricky to make clear but is not
particularly difficult, if it is done well.

Assessment: Students present some of the material (as far as possible
in areas that they are more comfortable with than the rest of the
class), they also do an end of term presentation of a project (the
idea is to get them to write code of some kind) and there is also a
written exam (which can be quite revealing).

Problems: It is very introductory but it has to be, given the varied
backgrounds of the students; the main goals are to give students from
the physical sciences some understanding of what the brain is like as
a physical system and to persuade the less-mathematically trained
neuroscience students that programming and modelling are not difficult
if you are clear about what you are trying to do.


FRANK (Arizona)

This past semester I taught a graduate seminar, ¨Computational
Cognitive Functions of the Prefrontal Cortex¨:

https://www.polis.arizona.edu/courseHomesite.do?course=PSYC_596F-1&semester=spring06

There were about 10 students.

This coming semester I will teach an undergraduate honors seminar on
computational cognitive neuroscience, with 20 students and one lecture
and one lab section a week. I will subsequently teach a grad-level
course on the same topic. Both these classes will use the O´Reilly &
Munakata (2000) textbook.


GOODHILL (U Queensland)

I currently teach part of a 3rd-year undergrad course in Mathematical
Biology, which this year had about 30 students. My part is loosely
based on parts of Dayan & Abbott, Theoretical Neuroscience, MIT Press,
2001, and focuses particularly on neural coding and unsupervised
learning.  Assessment is by computer assignments (in matlab) and
written final exam. The main problem is dealing with the diversity of
backgrounds of the students, which vary substantially along the 3
dimensions of maths, programming and neuroscience.