Pages

Monday, March 29, 2010

Random Signs of Time

As I was watching Ingrid Daubechies' talk given at UCL (in Belgium) on Independent component analysis and functional Magnetic Resonance Imaging (where she highlighted the connection between ICA algorithms and sparsity seeking algorithms) I had to recoil on the fact that she had to make a reference to CSI (31 minutes and 30 seconds into the talk): Yes, the same CSI that I referenced in Why Compressed Sensing is NOT a CSI "Enhance" technology ... yet ! and Surely You Must Be Joking Mr. Screenwriter. But I guess it was brought up for good. However, The CSI effect really has a dark side to it, and so it important that we keep making sure that what is feasible and what is not, is clearly understood by a large part of the population that is technically educated but is probably not up on their knowledge of the convergence bounds of the latest proximal algorithms. Case in point, I just saw this visitor on my log (please note the webpage that person read).



Ever since the blog has become a little more popular, I have gotten many "opportunities" including invitations to provide my name and address because I won the British Lottery but also I have been asked to provide reviews or be the editor of some new journals focused on compressive sensing. I am of two minds on the latter offers: On the one hand, I believe that the peer review process is good to weed out the truly awful papers. On the other hand, I cannot say I have published in this area nor that it is a good use of my time. I also believe that starting a journal solely focused on compressed sensing is the beginning of a long slide toward irrelevancy. I am not tenure track and have never been hence the road to funding is, as some of you know, a little overwhelming and justifies that I am very stingy on the time thingy. Finally, the blog is destined to be inspirational. The hard work is mostly done by most of you readers.

One could say that this is an easy way out but I stand to loose a whole lot more being ridiculed in public by you people for mis-stating something you say than by anonymously rejecting your paper. Furthermore, the style of my reviews would certainly compromise my anonymity. Pierre Vandergheynst has another take in his twitter stream, he furthered his thought in the following tweet. His sentiment echoes one of my recent op-ed.

This is all the more exciting to see that some of you decided to link to this blog for your courses or your book. The recent referrals I have seen in my log include:

Thanks guys !


After having watched Pierre Vandergheynst's presentation, I inquired about the need for his team to make a toolbox available so that one could use their CS imaging chip remotely at first and perform trade studies more efficiently. Pierre tweeted back:

it will be toolbox season at EPFL as we shoot at making our research more reproducible.

Outstanding! Not only will it be more reproducible, I also believe it will give them an edge as it will provide design engineers in start-ups or large companies a way to evaluate whether this is something for them. One way of dissiminating this type of "API" (besides advertizing it on the Compressive Sensing Hardware page) is to reach out to the 360 people who have joined the Compressive Sensing Study Group on LinkedIn. Let me know if you also want to be added to the Twitter list of folks that have an interest in Compressed Sensing.

In a different direction, Dick Lipton just wrote how Fast Matrix Product was discovered:
...One day he suddenly realized, if there was a way to multiply two by two matrices with {7} multiplications rather than {8}, there would be a sub-cubic recursive method for {n} by {n}. He immediately sat down and quickly discovered his now famous formulas. At least this is the story....

I like that story, mostly because sometimes the naive way of doing things is not the most efficient. What about if there was a Fast Matrix product equivalent to MMA17 ?

The talk I mentioned at the beginning of this entry was Independent component analysis and functional Magnetic Resonance Imaging by Ingrid Daubechies . The abstract of the talk reads:
Independent Component Analysis (ICA), a method for separating a mixture of different components into its constituents, has been proposed for a variety of different applications, including functional Magnetic Resonance Imaging (fMRI) of brain processes. The presentation summarizes the findings of several years of interaction between applied mathematicians and neuroscientists, expert in fMRI, concentrating on probing ICA methods for brain fMRI. This study raised questions, informed by mathematical considerations, that are investigated using numerical simulations and specially designed fMRI experiments. The intent was not to cast doubt on the successes of ICA methods for fMRI data analysis, but rather to understand the elements that determine the methods' success; this led us to a surprising result.

Bob Sturm has a similar question in acoustics.

In other news, the ONR funded Moeness Amin and Bijan Mobasseri at Villanova with $355,000 on compressive sensing projects. From the press release, I note the mind blowing statement:
Other venues include covert communications using biologically-generated emissions from marine mammals as cover signals,
This is not Imaging With Nature anymore, but rather Communicating With Nature :)

Marta Betcke at UCL (U.K) is leading a grant on compressive sensing CT scanning for security application.

The IEEE International Conference on Computational Photography (ICCP) is taking place right now at the MIT Media Lab. They have a twitter stream and a posterous site. They also have a live webcast channel. Ramesh Raskar continues to advertize for people to join his group:

We are actively looking for Graduate Students, MEng, PostDocs and UROPs starting Spring and Fall 2010. Please see here if you are interested.


Finally, a call for Green Computing that includes an interest in Compressed Sensing.

No comments:

Post a Comment