lidsblog

Spread the Wealth Around

Barack Obama to Joe "The Plumber" Wurzelbacher: I think when you spread the wealth around, it’s good for everybody.

I was out of the country last week and as always, was glad to hear the words "welcome home" from the immigration officer upon my return.  I came back to America, but apparently not to "real" America.  I participated in the IEEE International Workshop on Machine Learning for Signal Processing (MLSP), which was held in Mexico this year.  At the conference banquet, I sat between J. J. Remus of Duke University and Sergios Theodoridis of the University of Athens.  Remus, who is originally from Wasilla, Alaska (for certain a part of "real" America), had a poster on distance-weighted nearest neighbor classification.  Theodoridis gave a plenary talk and a regular oral presentation on using a sequence of projections onto convex sets in reproducing kernel Hilbert spaces to find feasible, but not necessarily optimized, solutions to problems such as finding classifiers. 

A paper on sparsity measures, which relates to the quotation at the top of this post, was presented by Niall Hurley and Scott Rickard.  Rickard had proposed six desiderata for sparsity measures in 2004, of which four were originally proposed by economists in the early part of the last century in the context of wealth inequity.  In the MLSP work, it is shown that among fifteen sparsity measures, including the popular ℓ₀ and ℓ₁, the Gini index alone satisfies the six desiderata.  Considering a vector of coefficient absolute values, where a large valued coefficient is analogous to a rich person, the six desiderata are:

1. Robin Hood.  Stealing from the rich and giving to the poor decreases sparsity.
2. Scaling.  Multiplying the wealth distribution by a constant does not change sparsity.
3. Rising Tide.  Adding a constant to each coefficient decreases sparsity.
4. Cloning.  If there is a twin population with identical wealth distribution, the sparsity in one population is the same as in the combination of the two.
5. Bill Gates.  As one individual becomes infinitely wealthy, the wealth distribution becomes as sparse as possible.
6. Babies.  Adding individuals with zero wealth increases sparsity.

According to the analogy, a sparse signal representation is clearly not an Obama signal representation.  I think I'll put in a request to change the title of a journal paper of mine to Palin Representation in Structured Dictionaries. 

Convex Optimization

Why can't things be like they used to be?  Pan Am and TWA?  Flying without quart-sized, clear, plastic, zip-top bags?  Boarding economy class from the back of the aircraft to the front?  As Pablo mentioned in the last lecture of 6.255, the newfangled zone boarding is the solution to an optimization problem to minimize boarding time.  I don't agree with the model used to describe the constraints and particularly the costs, but I don't doubt that zone boarding is the optimal feasible solution for the given model. 

I had the pleasure (or hassle) of zone boarding four times recently on the way to and on the way back from Las Vegas.  I wasn't there to emulate 21, or to lose a hundred dollars in ten minutes, but to attend the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP). 

The students of LIDS were there in force.  Myung Jin Choi & Venkat Chandrasekaran had a paper Maximum Entropy Relaxation for Multiscale Graphical Model Selection.  Dmitry Malioutov had a paper Compressed Sensing with Sequential Observations.  Mesrob Ohannessian had a paper A Turbo-Style Algorithm for Lexical Baseforms Estimation. Xiaomeng (Shirley) Shi had a paper Joint Base-Calling of Two DNA Sequences with Factor Graphs.  Vincent Tan had a paper Learning Max-Weight Discriminative Forests.  Lav Varshney & I had a paper Minimum Mean Bayes Risk Error Quantization of Prior Probabilities.

Aside from the Elvis impersonators and mimes, one interesting thing at the conference was a panel on convex optimization theory and practice in signal processing.  The main lesson that was imparted was that people working on signal processing applications really need to learn and use modern techniques of convex optimization, as well as collaborate with optimization theorists starting at the initial stages of problem formulation.  (One example I can point to where convex optimization theory has been used is Jason Johnson's Lagrangian relaxation work.) 

LIDS is uniquely positioned in this regard because we have both signal processors and optimizers within one lab.  Some members of LIDS are aeronautical engineers; now if we also had overhead bin designers, we could finally make aircraft boarding hassle-free. 

Carpal Tunnel Syndrome

First turning on the TV at the Hilton hotel in Houston, something most unexpected came on: a commercial advertising the Brown Procedure, an endoscopic treatment for carpal tunnel syndrome. 

I was in Houston on a junket visiting Shell International Exploration and Production's Bellaire Technology Center with Biz, Dahua, Dmitry, Hyun, Jin, Venkat, and Vincent alongside.  Back in the day when Bell Labs was Bell Labs, the Bellaire facility was the Bell Labs of the oil industry.  The mission that the eight of us chose to accept was learning how teams of geologists and geophysicists interpret data, primarily seismic data, to find reservoirs of hydrocarbons from which it might be profitable to extract oil and natural gas. 

Dahua summarizes his experience here.  Interestingly, since my eyes are now looking for it, I keep seeing articles about this stuff, e.g. these two articles in the December SIAM News and this article in the January IEEE Spectrum. 

There are no pools or rivers or streams of oil in the earth's crust.  Oil exists in the pore spaces of rocks such as sandstone.  The presence or absence of three things make or break a prospect: source, reservoir, and trap.  Source means that over geologic time, organic material transformed into oil through heat and pressure.  Oil is less dense than rock, so it tends to move towards the surface if possible.  Reservoir is a layer of the crust with pore space, a good example being a layer of sandstone.  Trap is a geometric arrangement such that the reservoir is surrounded on all sides except the bottom by impermeable layers such as those made of shale.  Trap geometries are often the result of faulting, i.e. the cracking and pushing up of some regions leaving other regions below. 

In some parts of the world, like under the Gulf of Mexico, the crust contains massive bodies of salt.  The salt behaves like toothpaste, getting squeezed and pinched by the weight of the surrounding rock.  Importantly, salt is also impermeable to oil. 

To image the crust, seismic data is collected.  A large amplitude acoustic pulse or chirp signal is input into the ground and the reflections are recorded.  Reflections occur from interfaces between different rock types such as sandstone-shale boundaries.  Reflections from the same spatial location are recorded many times because a single record has extremely poor signal to noise ratio.  The data that is collected for a spatial location is in time, not depth.  It must be migrated from time to depth taking the speed of sound in the materials below into account, an ill-posed and challenging problem.

With a seismic volume either in time or depth, the interpreter's laborious job is to pick out faults, salt bodies, and continuous layers of reservoir rock.  Then the task is to understand the depositional environments, geology, etc., identify prospects, and finally decide whether the risk to drill them is worth it.  (It costs somewhere around one hundred million dollars to drill an exploratory 'wildcat' well.) 

Shell has developed some automatic pickers that we got to play around with that somewhat reduce the laborious, repetitive nature of picking faults and events.  However, the tools are far from perfect.  Interpreters are still subject to much repetitive strain injury such as carpal tunnel syndrome.  Now that we have seen the procedure that interpreters go through and what open problems exist, we can contribute to the development of interpretation tools for geologists and geophysicists so that they can focus their time on the higher level, understanding-based tasks of interpretation rather than on the lower level, repetitive tasks. 

An eventual goal is to put Dr. Brown out of business and develop a full object recognition system, but that is a long way off. 

ISBI

Last weekend I was at the IEEE International Symposium on Biomedical Imaging in Arlington, Virginia.  The main reason I was there was to present research I did last summer in Paris, which was related to medical image segmentation using level-set methods.  It is not uncommon for LIDS students to spend time at other schools and research institutions or to do internships at companies over the summer.  Here are a few thoughts spurred on by the conference.

The opening plenary talk of the conference was delivered by Elias Zerhouni, the director of the National Institutes of Health, who apparently also delivered the commencement address at MIT the year before I got here.  You keep hearing it and hearing it, that life sciences are the thing for this century, and Zerhouni said nothing that would contradict that.  He talked about how medicine needs to become predictive, personalized, preemptive and participatory.  The idea being that it shouldn't be "wait until you're sick, and then come see me."  If you can look at your genome, cells, etc. now and see that you will develop a certain disease in twenty years, then you can try doing something about it now.  I wouldn't necessarily say that this is a completely new outlook on medicine, because many systems of medicine other than the presently prevailing one are holistic in nature and have much of the p's covered already, but then again, I am no doctor. 

Obtaining and using localized spatial and temporal biological information at different scales by imaging science is a key part of the transformation according to Zerhouni.  Imaging is not just about taking pretty pictures anymore -- it's about doing real science through quantitation.  One of the things distinguishing this conference from others is its ongoing theme of "from nano to macro," in that it brings together people doing medical imaging, e.g. radiology, and people doing biological imaging, e.g. microscopy. 

Biological imaging is a very different ballgame than medical imaging.  In medical imaging, everyone, i.e. the physicians, scientists, and engineers, pretty much knows the specific objective or problem.  For example there is a problem of segmenting the left ventricle of the heart, and it is clear what it is.  In biological imaging, scientists often don't even know what phenomenon they are looking for before they run an experiment.  Also, every laboratory's protocols and procedures are different. 

So how is a person involved in image processing, statistical signal processing, or machine learning supposed to get involved with the life sciences, especially biological imaging, which are supposedly the big thing?  As things stand now, my opinion is that the only way to get involved is by seeking out a biologist collaborator and working with them closely.  One big issue with biological imaging, again in my opinion, is that biologists are guarded with their data and laboratory results, inhibiting the free flow of information and stunting development on the image analysis side of things. 

If you are developing advanced algorithms for multi-target tracking, which many LIDS members and alumni are, and you can see their need in analyzing time-lapse microscopy, what are you to do?  Luckily at MIT, we can try to find someone across the street.  In general, I feel there is a need for some online forum that would enable biologists and those working on image analysis research to get connected.  Because developing large, standardized, publicly available collections of image sequence data seems counter to the culture of biological research, this seems like the way to go.

• Related to my previous entry, sparsity was on display at ISBI as well.  Andrew Bolstad and Fijoy Vadakkumpadan both presented papers (abstracts here and here) on magnetoencephalography that made use of sparsity regularization.  Both were aware of my master's research, which definitely feels good.  Michael Wakin proselytized for compressed sensing, and in the paper of Cao, Bouman, and Webb a technique for sparsely decomposing an orthonormal transformation was given. 
• There was an interesting problem hinted at by Henkelman in his plenary, namely that of classifying vasculature patterns into healthy and diseased categories.  In the liver and kidneys, for example, the blood vessels form a tree-structured pattern that is different individual to individual.  Tumors and other causes of ill health alter vasculature. 
• Liang gave a rousing tribute to the departed Nobel laureate Paul Lauterbur. 
• My officemate Pat Kreidl has some qualms with the great expenditure associated with conferences, which I agree with to a certain extent.  However they are a great opportunity to meet with and talk to people and to get exposed to different ideas.