?

Log in

Technical Ramblings
 
[Most Recent Entries] [Calendar View] [Friends]

Below are the 5 most recent journal entries recorded in Aprotim's Thesis' LiveJournal:

Monday, October 18th, 2004
2:30 am
[herbie]
Things to do (not necessarily thesis, but research related):

  • make figures for paper

  • write common visualization functions for matlab

  • work with Ben to allow NeuroJet scripts to include functions
  • write NeuroJet function to do MetaJet things</li>
  • add matfile capabilities into NeuroJet

  • my thesis

Wednesday, July 21st, 2004
2:29 am
[herbie]
So after being linked here, I figure I should update - I have gotten to the point that I can more or less do everything I want automatically as regards sweeping - I managed to just generate graphs within minutes of being asked (sometimes... sometimes hours or days - but after very little personal effort, just a lot of chugging on simulations).  Last step - integrate data aggregation/analysis into the sweeping. (Right now, I generate a lot of raw data, then I run a hacked together script to analyze and digest, and another to further compress and graph - the first and last steps are automated, in many cases, but the middle is just always hackish and not very generalized.)   The way I see it, I can integrate some matlab work into the sweeping, or I can integrate some analysis into my .nj (Neurojet) scripts.   Most of the functionality I need is there, but not all, so i may have to spend a little time working on a patch for NJ and giving it to ben for consideration - hopefully this would be something very quick and dirty and not take too much time away from the neuroscience.

Speaking of which - we've done some very interesting sweeps of NMDA-rule competitive networks doing trace conditioning over various intervals - there is a definite linearity of anticipation vs. trace interval initially.  There is also a phase transition - or perhaps better described as a threshold of complete collapse (I'll post graphs forthwith), and potentially some asymptotic behavior at diffrent threshold criteria for prediction.  Futhermore, I have some initial data that suggests that by reducing network connectivity (that is, percentage connectivity, not absolute number of connections), that phase transition can be delayed.  Additionally, i have one sweep that shows a distinct reduction of the slope of the initial linearity with a larger network.  I'm investigation both these effects.

That is I need to vary:
  • connectivity
  • network size
Additionally, Chip asked me to investigate the effects of changing:
  • alpha (NMDA-rule time constant)
  • me (I'm not sure what kind of effect he expects to see here)
Friday, June 4th, 2004
11:49 am
[herbie]
So, i have lots of data, and little ability to actually write it up. I can't figure out what to say, except "here's my data - I hope you like it."

On the other hand, my sweeping tool is getting steadily more sophisticated, I have the ability to easily sweep up or down any given macro in your script (I have to decide whehter I want to keep doing this macro-style, or make it so that you can modify real runit NeuroJet variables.) It can also resume old sessions, start at an arbitrary seed index (it has fixed seeds to make it easier), and I'm currently working on making it capable of grabbing multiple output files, including globbed ones.

Next step in that regard is to finish making the multiple files work, make sure that macro reordering works, and then create general purpose graphing scripts that will simply ask which variables to graph over, which to hold constant, and will generate matlab code to do it.

In other news - I'm setting up a CVS server, and I'm hoping to do a full cleanup of the runit/punit code along with Ben for to make it ready to be maintained of NeuroJet.
Friday, May 21st, 2004
3:09 am
[herbie]
Actual Research/thesis stuff
So - today was a sick day. Blech. However - yesterday I set up rr (rollsroyce) with Debian as a filebox/workstation for Aleks, cleared out desk space in the back room, moved cortex in there, and settled myself in there (tidied it up a bit), cleaned up my old desk (of Kurt Mitman's stuff) and put rr there for Aleks to use. (Today Aleks got to try to get X working to his liking - i.e. at greater than 800x600).

I'm relatively happy with my sweeps - not because the data's that great, but because my sweeping tool is getting to the point that I can use it for anything. And with a little tweaking and a PVM refresher, it will be ready to be thrown on the Turtles (Alphas) once I figure out how to get them running.

I need to write, and I've started, but it all seems like "here are shapes - they are interesting, maybe look at them and let me know". There's no solid trend I can see across simulations, so it's hard to extract results. I'd like to get into the innards of RUNIT, and try to work on my topology problem, but I guess walk before I run. Besides, a whole lot of prep work needs to be done if Dave, Ben, Shaft and I are going to be working on runit, like getting a CVS repository up and running, and fixing flawed assumptions. Also, I get the distinct feeling that the runit code is horribly monolithic, making it hard to deal with runit/punit compatibilities. It may be that it's necessary because there are assumptions you can make to make runit faster which you can't make for punit and vice-versa, but I think while the heart of runit seems to be ok (just by assumption - I haven't really looked), we may need to make it easier to maintain.

Besides that, in order to get my topology project off the ground, I need to do some serious research/thinking into how to specify connection topologies in relatively arbitrary ways. (For example - imagine the visual cortex - you need to map each retinal ganglial cell to simple/complex cells in a particular way, where the synaptic pattern is of a certain form, and the topology of neighboring complex cells is fairly simple, but needs to be expressed in a way in which it can be expanded automatically to create a large region from a simple specification. It sounds easier now that I've written it down, but I still don't know how to do it.)

All this, and besides it looks like I'll have to finish up this TC paper after Dave leaves... Hmm... things to think about, but right now, bed time.
Wednesday, April 21st, 2004
4:59 pm
[herbie]
First Post!
This is my journal for fourth-year thesis-type things. I hope topost a lot to this, as a way to document progress. Fun! (I will also, hopefully be keeping a paper journal.)

Huzzah.
About LiveJournal.com