|
In this assignment, you will implement or modify an existing Topographica simulation to model a novel phenomenon or to add a novel mechanism, writing up your results in the form of a short research paper. In this way you will get experience with selecting research topics, locating and understanding scholarly neuroscientific and computational literature, and presenting scientific results. The goal is to end the course in a good position to start constructing novel, feasible, and publishable models of the visual cortex. The assignment is deliberately open ended, and you should be careful to keep in mind the overall fraction of your course mark that this assignment represents (30%) so that you can spend your time accordingly.
You are strongly encouraged to work in pairs of your choosing, in which case you should submit one copy of the assignment, listing your partner, and the mark will be allocated equally to each. Please note, however, that both partners must contribute equally, and that pairs are expected to work on most parts of the project together, helping each other understand and figure things out, NOT dividing up the sections to be done separately. The scope or quality of a joint project will be expected to be somewhat more than for a single project.
First, you should choose an appropriate topic and research plan. Your plan should consist of novel, concrete computational experiments that you can do to build upon published peer-reviewed work on models or theories of topographic maps, or experimental data relevant to topographic map formation or function. There are three equally important constraints to consider: your research questions must be possible to answer computationally, the results (if successful) should be novel (not directly replicating existing work), and the project must be feasible in the short time available (which means that the work is likely to be incremental, which is fine). Unless you choose the default dichromatic vision option, you must discuss the general topic area and research plan with me (Jim) before starting work in earnest, in part so that I can ensure that it is novel and feasible (and give my opinion on how interesting it is :-).
One obvious approach would be to choose a model already implemented in Topographica, and extend or modify it in an interesting and novel way. However, developing a small model from scratch or reimplementing some other published model is also fine, as long as you are not overly ambitious.
In either case, some possible questions to consider are described in the Future Work chapter of the CMVC text, but note that many of those have now been done by people here and elsewhere, so it is vital to check with me first. Some that are still open include:
Add suitable additional laminae in V1 (beyond L4 + L2/3) or additional cell types, with specific circuitry motivated by experimental results.
Implement a basic model of invariant object recognition (using principles from HMAX or VisNet, but in a topographically organized region); see CMVC 17.2.11.
Implement learning in push–pull afferent connections to V1 (where inhibitory feedforward connections should strengthen when the neurons are anti-correlated); see CMVC 17.1.2, Troyer, Miller, et al. J. Neurosci 1989, and Kayser and Miller (2002). Ideally, the same model could then be used for both LGN and V1 neurons; right now LGN neurons have feedforward inhibition but V1 neurons do not.
Replicate and explain species differences in topographic maps, such as for joint OR/OD maps; see CMVC 17.2.2 (and published work from other models, e.g. from Wolf and Geisel, and possibly the elastic net).
Develop a more detailed model of the process of OD map development — effects of prenatal retinal waves followed by postnatal visual experience; see CMVC 17.2.3 and Jegelka et al. 2005. May need to implement layer 2/3 on top of layer 4, with timing differences, to explain the data.
Derive and demonstrate an interesting analytical result about topographic map models (e.g. proving aspects of them optimal or suboptimal according to information theory or probability theory); see CMVC 17.3.1.
Adapt the LISSOM LM or LMS color simulations to model dichromatic LS color in a non-primate species such as ferret; see work from Johnson and Fitzpatrick.
Adapt the LISSOM direction map simulations to develop direction selectivity via intracortical delays rather than delays in the afferent projections.
Starting from the retina, develop LGN-like cells from self-organization, as in various previous LGN/RGC models (Haith, Eglen, etc.) but with learning rule similar to GCAL; will need learning rule for feedforward inhibition (above).
Please bear in mind that you are not required to get "good" or even interesting results, as long as you make a good-faith effort to evaluate your hypotheses. Your report (below) will include whatever results you obtained, whether or not they matched your expectations or establish the point you set out to establish.
Introduction/Background/Literature Review: What you are going to look at, and why that is novel and interesting. This section or sections must include references to specific scholarly, peer-reviewed papers (e.g. from pubmed.gov) on which you based your work, including experimental work, and must explain exactly how your work relates to them.
Methods/Architecture: How your approach differs from a specific earlier computational paper or papers. If your model is simple, you can report all of it; otherwise you need only report precisely how it differs from previously published work (and why).
Results: Figures, analysis, and text demonstrating what you found.
Discussion/Analysis/Future Work: Interesting aspects of the results, their significance, how they relate to biological or other prior data, and possibilities for further work.
Conclusion: What the reader should take away from this study.
Bibliography: Any information that you use that is not from the course material or your own experience must be cited in the text, typically using (Author, Year) format. The complete bibliographic info (including everything stated in my writing tips) must then appear in the bibliography. You can use the CMVC book Bibliography database for citation information for any paper cited in the CMVC text.
If your results support the hypothesis you were testing, then the report should convince the reader of the truth of that hypothesis. If things do not work out as you planned, you just need to convince the reader that you had a reasonable approach, and that you put it into place properly.
Your report will almost certainly improve if you consider my list of writing tips. Also, when preparing output from the Topographica simulator, please keep these tips on generating high-quality figures in mind.
Your work must be submitted by the deadline, using the
submit command on Informatics DICE machines (type
man submit for more details). Your work should be in the
form of one plain PDF file named report.pdf, plus one or more .ty and/or
.py files with your code.
Standard late policies will apply as described in your course guide, namely that late coursework is not accepted without good reason, which must be discussed with the ITO, not the lecturer. Similarly, standard academic misconduct policies apply as described in the University guidelines; in particular; students must clearly label any aspect of their submission that is not their own independent work (or that of their named partner).
Example of submit command:
submit cnv 2 report.pdf model.ty components.py
/group/teaching/cnv/topographica/topographica -a \
-c "run_batch('asst2.ty',analysis_fn=default_analysis_function,\
times=[100,1000,2500],cortex_density=30)" -c "save_snapshot()"
This runs the default_analysis_function (defined in
topo/commands/basic.py) at each of the specified simulation times.
You can copy that function to your own .ty file, change the function's
name, and add any code inside it that you want to run repeatedly for
the simulation.
Any parameter that you add at the end of the run_batch
command, like cortex_density=30, will be set before
running your .ty file, and will be put into the filename
of the output directory. You can use this mechanism to keep track of
the various configurations that you run. For each parameter you want
to control like this, just put add a parameter declaration for your
newvar like the one for retina_density at
the top of the file, and then put p.newvar wherever you
want to use the value.
For this assignment, you will need to convince me that you can do novel research in computational neuroscience, clearly conveying what you did, why you expected it to be significant or interesting, and what the results were (whether good or bad). Remember that I can only judge what is actually submitted, so you should make sure that the files you submit make it clear that you have done what the assignment requested, and that you have thought about what each part means and represents.
Submissions must be in PDF except by prior approval with the instructor. I can be sure to be able to read PDF; others like .doc or .sxw have a certain probability of working properly, but the probability is far from 1.0. Naming the files as I suggest will make my job a lot easier, because I will be able to see exactly what you are submitting.
If you need a citation for GCAL in your report, you can use:
@TechReport{law:uetr11,
author = "Judith S. Law and Jan Antolik and James A. Bednar",
title = "Mechanisms for Stable and Robust Development of
Orientation Maps and Receptive Fields",
institution = "School of Informatics, The University of Edinburgh",
year = 2011,
note = "EDI-INF-RR-1404",
url = "http://www.inf.ed.ac.uk/publications/report/1404.html",
}
Last update: assignment2project.html,v 1.4 2014/03/06 03:38:10 jbednar Exp
|
Informatics Forum, 10 Crichton Street, Edinburgh, EH8 9AB, Scotland, UK
Tel: +44 131 651 5661, Fax: +44 131 651 1426, E-mail: school-office@inf.ed.ac.uk Please contact our webadmin with any comments or corrections. Logging and Cookies Unless explicitly stated otherwise, all material is copyright © The University of Edinburgh |
