Wednesday, March 18, 2009

So what am I (Bob) doing out here?

For you people back home who are wondering what I am doing out here in Paris, this note is for you! The question has usually gone, "So, Bob. What exactly will you be doing in Paris?" "I will be doing research." "I see. Like looking through encyclopedias and writing book reports?" "Well, not exactly. I will be working with a professor who specializes in my discipline." "What discipline is that?" "I am glad you asked. Let me tell you with the help of some helpful pictures."

The title of my post-doctorate proposal can be seen above: Applying Overcomplete Methods to Scalable Audio Indexing, Search, and Retrieval. Since submitting this proposal in January 2008, and completing my dissertation in the meantime, many things have changed. For instance, I would now title this work: Applying Sparse Approximation to the Efficient Description and Indexing of Audio Data. (What is an "overcomplete method" anyway!? ha ha.) Furthermore, we won't worry about scalability until some later time when we have tackled some important questions, like:
  1. Can we effectively deal with the shift-variance of greedy pursuits? A sparse approximation generated by greedy pursuits of a signal can change dramatically with even a slight time-shift in the signal, or transforming a signal with an allpass filter -- even one with a single pole. The best thing would be to find a way to represent a signal in a sparse way that is independent of the time-domain waveform.
  2. Given two sparse approximations of signals, how may we compute the distance between them, or their content, such that meaningful (or at least useful) comparisons are made?
"So what is your day like then?" Well, let me use another helpful picture to give you an idea.

That paper, entitled "Gradient Pursuits," is one that I am currently studying. You see, I find and read research articles, conference papers, Ph. D. dissertations, even my own work, that are relevant to our big questions. I read them and I spend some time summarizing each one, framing their work in terms of ours, seeing what can be useful, and what can be improved. The most satisfying part is making an "X" in the top-left corner with a date, and sticking it in a pile of "has-reads." Seeing this pile grow is like looking at the marks on a wall denoting my height.

There is also the copious amounts of tea and little biscuits. I think I must have about 4 to 5 cups a day of this wonderful fuel -- from orange pekoe, to Barry's Irish blend, to lapsang souchong, terry souchong, pu-erh, French blue, etc. All of these provide wonderful pauses between my clauses.

There you have it. Research. French style. I think it will be hard to return to the states, because I won't be fitting in no aeroplane.

No comments:

Post a Comment