What the SCOPE 2003-4 program did for meDinoj SurendranThe original form of this report was written and submitted to Leo Kadanoff, the PI of SCOPE, in May 2004. The sections in italics were added in August 2004. SCOPE played a large part for me in my development as a human, as an apprentice researcher, and in considering science popularization as a career option. Much of the credit for this goes to the coaches of the Astronomical Sciences division of SCOPE, Randy Landsberg (Director of Outreach, Kavli Institute for Cosmological Physics and UChicago Astronomy & Astrophysics Department) and Mark SubbaRao (Research Scientist, UChicago Astro Department, and Astronomer, Adler Planetarium & Astronomy Museum). I learnt how to use Partiview, a 4d visualization tool for displaying galaxies found by the Sloan Digital Sky Survey, and realized it was close to perfect as a data exploration tool in machine learning. I made a few demos (using data from handwriting and speech recognition) and presented them to several people, including machine learning researchers (Partha Niyogi, Misha Belkin), other graduate students (at our departmental graduate student seminar), and the Partiview community. This got Stuart Levy, the NCSA researcher who wrote Partiview, very interested, and he has been willing to add features that I've requested, such as commands that make Partiview work much better on a GeoWall, and commands to make it more useful for machine learning researchers. I aim to combine Partiview and various dimensionality reduction algorithms into a new tool for visualizing high dimensionality data. For the moment, I have to submit proposals for presenting these at the Information Visualization Conference in Austin TX in October 2004 (update: interactive poster, with Stuart Levy, accepted), and the Neural Information Processing Systems conference in Vancouver BC (this interactive demo has since been submitted) in December 2004. I'll also show it off at the Machine Learning Summer School on Berder Island in France in September 2004 (student presentation abstract submitted), a workshop for which 80 out of 200 applicants were accepted; at least one of my referees was probably more impressed by my visualization-making abilities than my research abilities (translation: I may not have been accepted if I hadn't gone through SCOPE). The strange thing is that none of the work I'm combining is actually new. (July update: Maybe it is somewhat new, since my digits demo got through one round of the NSF's Science and Engineering Visualization Challenge.) Certainly many people in machine learning still don't appreciate the importance of visualizing data. Anyway, fun stuff happens where disciplines meet. I've also used Partiview to help other graduate students (e.g Leandro Cortes, a CS student working on biology problems with Eric Schwartz) visualize his 3d data, and have begun working with Adriana Iamnitchi (Visiting Assistant Professor at Duke) on an exploratory data analysis project with grid computing. Whether it will actually work is another matter, we're still in the proof of concept stage. Update: it doesnt appear to have worked out... it's certainly in the backburner stage. However, I've also sinced produced some useful visualizations for another Foster student, Matei Ripeanu. All the above was not expected when I joined SCOPE; in fact when I began SCOPE in September 2003 I strongly suspected I would be kissing goodbye to my research career. That said, I have said goodbye to certain parts of my research career; there's definitely been a shift of my research areas, a shift that SCOPE catalyzed and that I do not regret. Rather, that I do not regret much - there's always regret when one turns away from an interesting unfinished problem, even if it is to look at more interesting problems. One of the best aspects of SCOPE is that I've met cool people through it, people I wouldn't have met otherwise. Not just the coaches and other interns (some of whom I get along very well with and will keep in touch for the foreseeable future) but also grad students, postdocs (Brian Humensky, Alan Calder) and faculty (Josh Frieman, Lucy Fortson) at the astronomy & physics departments here, SciTech (Sammy Landers), Adler (Jose Selgado, Geza Gyuk), DePaul (Jim Sweitzer). I've also made email contact with several people elsewhere e.g. Brian Abbott (AMNH), Stuart Levy (NCSA), Hiranya Peierls (Princeton), Selden Ball (Cornell). The resulting sense of perspective has been stellar, compared to what I had before. Not that it's anywhere near enough yet. SCOPE has also given me a sense of what it's like to work with other people. During the middle of the winter quarter (and probably during the middle of the night), when I felt (wrongly or rightly) I was the only intern doing any work and was quite frustrated, I came up with the following lessons:
In retrospect, I would only rephrase the first lesson, to something less paranoid. That's because teamwork did happen, particularly in the Spring Quarter, and has been very cool when it's happened. Within the group, I worked with :
I also worked with Mark and Randy on giving several demos, and getting our GeoWall products used in an undergraduate NATSCI 102 class taught by Rocky Kolb. The result of that was that our products proved to be robust enough to use by someone other than us, and a useful teaching tool in the hands of Brian Wilhite, a teaching assistant for the course who was familiar with Sloan Digital Sky Survey data and therefore appreciated it a lot more. I also worked with people outside the group (mostly in the Winter and Spring quarters):
I've probably missed out a few names. Clearly, SCOPE has done a lot for me, and I'm indebted to it for that. Much of the credit for this goes to Randy Landsberg, who knows how to motivate people worth motivating, how to put them in contact with other people, and whose views about how a museum designer should be trained (learn by doing) are completely in line with mine. And I'm speaking as someone who knows he has only scraped the surface of what the museum design process involves. And to Mark SubbaRao, who introduced me to Partiview - which, as indicated above, has changed the course of my research career - and converting between different astronomical formats. Both have provided support in other ways too numerous to name. Future plans: I am torn between choosing between research and science popularization, and will probably try to do both for as long as possible. I certainly have enough to do for the next year, like getting a doctorate, and making pretty pictures for research and outreach purposes. For the immediate moment, i.e. summer 2004, I'll be working with Mark and Randy on making more astronomy-related visuals for the KICP and Adler. That should be fun, and a lot of work. But SCOPE's been a lot of work too. Postscript: August 2004My work with SCOPE led to a summer job making visuals for a short course run by the Kavli Institute for Cosmological Physics. This has primarily involved visualizing cosmic ray showers the Pierre Auger Observatory and VERITAS, and maintaining the Cosmus page. And collaborating with FLASH scientists on getting their supernova (etc) simulations on a GeoWall. It also resulted in an invitation to present some of the visuals Mark and I had made to scientists and educators attending a Physics Education and Outreach workshop at the Aspen Center for Physics in July 2004. |