Sunday, December 13, 2009

Switching gears

I posted a few times based on my encounters in brain image analysis at the Waisman center. Today I wanted to post a sample demo script on reading output from DTI_TK in MATLAB. Then I realized I will wrap the demo script in a meta post about how important it is to be able to switch gears (if one wants to reduce overhead in interdisciplinary research) not only in terms of conceptualizations but simple things like being able to use multiple platforms and software packages. Since I joined Waisman, I got more comfortable with MAC, LINUX etc. I am no longer just a WINDOWS person although I still cannot trade WINDOWS for any other platform. I plan to release some tools for CAMINO TRACKVIS interoperability as well.

See, all research in interdisciplinary at some level and needs ability to switch gears but applied research just has higher demand (for doing noticeable (even locally) research) on the gears and needs sufficient investments in greasing it properly. I worked on projects in collaboration with Psychology departments before while I was at Temple. I interacted with "psychologists" working on computer vision problems. There the research is aimed at hypothesizing human perception based on Gestalt psychology (and here's the kicker) eventually leading to machines with perception. While working with psychologists in neuroscience like Kim the results should eventually lead to interpretations of human behavior for clinical purposes. Both these objective functions have quite different properties!

Most of my palpable research experience has been in coming up with heuristics motivated by Gestalt principles and apply blackbox methods from simulation based statistics, specifically particle filters for perceptual grouping and robot mapping problems. More lately I have been working with passionate young Assistant Professor,Vikas Singh whose interests are actually in applying and analyzing techniques from optimization theory and machine learning. This is opening up a lot of opportunities for me to actually start think and work on actually analyzing the efficiency/complexity of heuristics. I am seriously hoping to build some skills in "smoothed analysis". Then I need to be able to switch gears from thinking like complexity analyst, to psychologists, to being software engineer (one of my key skill-strengths). Anyways I will blog more specifics on that when I have some real progress in that line. Without further wrapping I will present what I originally intended to present that is a simple demo script to read output from DTI_TK in MATLAB.

You can download the package from here and run demo.m. I am not explaining the details because what I am offering is possible grease into the gear of DTI processing (to save some annoyance) not building the gear. The actual gear itself can be built pretty nicely using the documentation on the DTI_TK website itself.

No comments: