January 2011 Archives

I am happy to report that our massive revision has been accepted with minor revision! We are almost converged.

Paper deadline coming up soon: EUSIPCO 2011!

And I have learned of this interesting pattern recognition competition for music signals. In the music instrument recognition task, the authors have decided that features extracted from 40 ms segments are the best. There is already plenty of evidence that making such a naive assumption will limit classifier performance in such a task. For the music genre classification task, they have computed MPEG-7 features from 20 segments of unknown length. Strange. Still, this might be a pretty cool exercise to motivate the relevant portion of my class on artificial intelligence.
A few of my musical compositions have been selected to accompany a film being produced about the Nordic Game Jam 2011, which occurs this weekend at Copenhagen University. Woo who! A funny thing to mention is that I composed the "Introduction to the Logic Opera" from the solution to a logic puzzle. The circles in the grid specified notes and durations and what not. In fact, I was going to compose the entire opera (to be performed in a local café in Boulder, Colorado back in the late 90's) from this puzzle, including characters, their actions, the plot, etc., and the audience was going to attempt to solve the puzzle. It was to be a very "Robert Ashley affair," with lots of mysterious happenings, embedded performers, and beautiful what nots.

Alas, more than 13 years later the opera remains unfinished, except for "The Love Song" (again, composed from the logic puzzle, and in 5! and that is me "singing"), and a sketch of one of the scenes. Maybe some day I will return.
Hello, and welcome to Paper of the Day (Po'D): Performance Limits of Matching Pursuit Algorithms Edition. Today's paper addresses the very problem I was mystified by before the break: Y. Jin and B. D. Rao, "Performance Limits of Matching Pursuit Algorithm," IEEE Int. Symp. Info. Theory, Toronto, Canada, pp. 2444-2448, July 208. For example, see my results here, and here, and here, and here, and here, and here. In particular, the last link there shows an image of how horribly OMP recovers sparse signals with equiprobable Bernoulli distribution.

Xavier Serra is awarded an ERC Advanced Grant

| No Comments
Congratulations to Xavier Serra, Director of the Music Technology Group of the Universitat Pompeu Fabra, who has been awarded an ERC Advanced Grant to carry out a project related to multiculturalism and music technology. The Advanced grants of the European Research Council promote innovative, 'blue sky' research, and are open to top scientist from any discipline. The project CompMusic by Xavier Serra has been funded with 2.5 million Euros and will last 5 years. A brief summary of the project goals can be found at the MTG website.

Path-finding back to Copenhagen

| No Comments
I leave for home in only a few short days. And then it will be several weeks' worth of Po'Ds on which to catch up during my time-zone-adjustment period.

Among my scattered papers in my office is one that appears to fuse orthogonal matching pursuit (OMP) with the A* search algorithm. This paper caught my eye not only because of the OMP bit, but also because just prior I had taught some path-finding, which is where and when I first heard of A*. Well, now that I will be teaching in only a few weeks time a 10-lecture course on artificial intelligence, I have been delving into this topic, and programming my own fun MATLAB demonstrations (to be reproduced by the students, so no demo code here yet):

pathfinding.jpg Above we see paths from a start to a finish found by Dijkstra's algorithm, A*, and greedy best-first (left to right). The last two are instances of what are called "informed search," which just means that the algorithm chooses which nodes to explore further based on some guess about the distance yet to travel is used, a.k.a. a "heuristic." A* uses both the cost to get to the current node and the heuristic. The greedy best-first approach just uses the heuristic. Dijkstra's algorithm is an instance of "uninformed search," which just explores first the nodes with the cheapest cost thus far without guessing about the distance to the finish. Both Dijkstra's algorithm and A* are guaranteed to find the least cost path when there is one. We have no such guarantee with the greedy best-first search. However, we can see above that the greedy best-first search has the smallest "fill" of all --- it explores far fewer nodes than the others, and thus gives a reasonable path with a small amount of computation. A* has much less fill than Dijkstra's, but only finds one least cost path from start to finish, while Dijkstra's finds many least cost paths from the start to several nodes including the goal. (Dijkstra's algorithm, or a variant, is used in the IP layer of the 5-layer Internet protocol stack for packet routing.)

I can see very clearly how finding a sparse representation of a signal in some dictionary can be formulated as finding a least cost path through a set of nodes (atoms in a dictionary) starting with the signal and ending at zero norm residual, or within some radius of it to accommodate noise. Some members of the family of matching pursuits (MP, OMP, OLS) are just greedy best-first search where the heuristic is defined as magnitude projections of atoms (nodes) on the residual. So I wonder if keeping track of the cost, as done in Dijkstra's and A*, is just regularization, or an instance of a relaxed greedy algorithm?

One more week

| No Comments
in sunny Southern California, and then it is back to the dark cold of Copenhagen. Actually, even though it is in the teens here (and could get up to 24), I am a little on the chilly side.

Hey look, the back of my mug made the Glendale Newspress (at 7 o'clock). I had the great opportunity to observe and talk with several talented and motivated high school students about ethics and leadership. It was inspiring to see so many young people interested in becoming better people all around. One thing that struck me was the intense pressure all of them feel about being the best at everything in order to get into the best university. "Even the state schools won't even look at my transcript if I don't have a 4.0, or am in the top 1% of my class," many said. Some of the students admitted to taking 5-6 advanced placement (AP) courses (which give one partial college credit), while at the same time participating in several sports, and association responsibilities. How does anyone like that have a well-rounded high school experience?

Education is big business in the US, and with all the advertisements of higher education and their tone, it is no wonder many people have the idea that getting a degree entitles one to a job; and if said job does not appear, then take legal action. People are led to believe everyone must go to university in order to be financially successful. Even certain subjects are more favored than others, and with the recent $500 million cut from the UC budget, we will see which subjects are worthwhile and which are not. (Is there much patentable work in the humanities?) College preparatory courses are also big business, promising to give one a competitive advantage by teaching to the test, which I think describes the state of lower education of the entire US with the "No Child Left Behind" act. Money is so much of an obsession in the US culture that there is scant room for anything else, such as ethics, or knowledge for knowledge's sake. What about getting an education because it makes one more informed about the world, one's place in it, and because one's chosen subject is fascinating? How about because it is fun and challenging? It is not for everyone; and I would also say it is not for anyone wishing to become a billionaire based on the vanishingly small frequency of college graduates that have become billionaires. But hey, people still play the lottery.

I was asked at lunch and after by several people about my life in Denmark as an assistant professor. (And throughout the day I was asked, "Did you come all the way from Denmark for this???") After saying I was very happy in Denmark, with lots of time to work with interesting students and colleagues and topics, with free health care, free school, six weeks of paid vacation, etc. etc. etc., they asked, "But you pay really high taxes right?" Yes, and so what? My life ain't about money, and I am not living under a bridge heated by a fire in an oil drum. The Danish government is one of the least corrupt in the world, and bribery is extremely rare. Now, if I can just get the language down!
Preparations for my course (Artificial Intelligence) are underway full steam ahead! Below you can see boids wandering, seeking, and fleeing moving targets. This is fun!

Blog Roll

About this Archive

This page is an archive of entries from January 2011 listed from newest to oldest.

December 2010 is the previous archive.

February 2011 is the next archive.

Find recent content on the main index or look in the archives to find all content.