![[Home]](/ailab/people/andel/pics/NeuroWiki.png)
my homepage is at http://purl.net/net/bshanks.
i am a second-year graduate student in computational neurobiology at UCSD. Also, i was one of the founders of NeuroWiki.
recently, i did some thinking about a broad survey of the various ways that graphical structures could compute, including some unconventional ideas. Also, about how computing with synchronization might be modeled without differential equations. If you are interested in hearing more, drop me a line and I’ll send you an overview/brainstorm-ish writeup that I did for a class project.
right now i’m doing a lab rotation at UCSD in Prof. Henry Abarbanel‘s lab. i just started a few days ago, and i don’t know exactly what I’ll be doing, but it’ll involve modeling BirdSong in the ZebraFinch.
right now i’m doing a lab rotation at UCSD in Prof. Gary Cottrell‘s lab. i’m working to “tie up loose ends” with a model of the LeechSwimCPG that a cool grad student here, Adam Taylor, made. Here’s the paper on (some of) his work on this:
So Adam Taylor found data for 9 cells in the LeechSwimCPG. He made up some simple equations to try and model their dynamics. Free parameters in these equations include the synaptic strengths between the cells as well as some of their internal settings. Then he used the data to fit these parameters (using a two stage process; first, make some assumptions about the network that allow you to solve for the free parameters in the model. Now, use those settings as a starting point, and do a gradient descent search for better fitting parameter’s using Pearlmutter’s AIWiki:BackPropagationThroughTime algorithm).
The resulting network fit the data pretty well and oscillated, but it was very sensitive to small changes in the parameters, and it was not constrained to mimic the actual connectivity of the real LeechSwimCPG. So my project is to use a few ideas that Adam Taylor and Gary Cottrell had to see if we can make the network mimic the actual connectivity and make it more robust.
Last week I was looking for a model with more realistic connectivity. The procedure used above doesn’t seem to be finding working models when connectivity is greatly restricted, however, if you greatly restrict connectivity and then add in certain connections, it works. And if you take loose connectivity constraints and delete those same connections, it fails. This suggests that those connections are there. An alternate explanation is that we do not have all the cells in the CPG. Friesen, the top researcher on this system, speculates that at least one cell is missing based on physiological data (we call it “cell X”.
(much later).. okay, the summer’s over and I don’t have much time to write at this minute. Basically, we got the stuff to work with realistic connectivity, but the network was still too sensitive to perturbing parameters (including parameters like connection strength that should be at least a little insensitive). We hope at some undetermined future date to implement a “simple” hodgekin-huxley model, and expect that this will be much more insensitive to parameter perturbation.
-- BayleShanks
personal neuro to-do list:
to read for proj:
Leif’s citations:
Abarbanel HD, Huerta R, Rabinovich MI. Dynamical model of long-term synaptic plasticity. Proc Natl Acad Sci U S A. 2002 Jul 23;99(15):10132-7. Whitehead A, Rabinovich MI, Huerta R, Zhigulin VP, Abarbanel HD. Dynamical synaptic plasticity: a model and connection to some experiments. Biol Cybern. 2003 Mar;88(3):229-35. Abarbanel HD, Gibb L, Huerta R, Rabinovich MI. Biophysical model of synaptic plasticity dynamics. Biol Cybern. 2003 Sep;89(3):214-26.
See also NeuroWikiTodo
Flavio’s webpage: www.frohlich.ch.vu
my UserId? here is 1002.
bookmarks for me:
Hey Bayle. Good idea to take a look at this. It’s my second favorite organ (W. Allen) – MattisManzel
Luigi Bertuzzi is contacting Peter Small for a colaborative translation of his works on stigmergy and stuff, seems to relate. http://www.stigmergicsystems.com/nb/culture.html anyhow. – MattisManzel