Sunday, July 06, 2014

Sunday Morning Insight: Mathematical and Algorithmic Analysis of Network and Biological Data & Open problems from the 2014 Bertinoro Workshop on Sublinear Algorithms

Back in 2012, in predicting the future: The Steamrollers, we mentioned how sensors, especially in genomics would change our fields. On Friday, based on one of this earlier observation and continuous monitoring of the technologies involved, we mentioned that we are potentially witnessing a Second Inflection Point in Genome Sequencing ? and then wondered what it would entail briefly. In the second part of that prediction, we envisionned some of the algorithmic tools that might be used to make sense of the data coming out of the steamrollers (Predicting the Future: Randomness and Parsimony). Here is one piece of that puzzle: Charalampos (Babis) Tsourakakis's thesis entitled Mathematical and Algorithmic Analysis of Network and Biological Data that provides a year old view on the subject. For more, his arxiv preprints can be found here. You might also be interested in reading his tutorial on Algorithmic techniques for modeling and mining large graphs (AMAzING). From the thesis abstract:

"....Our contributions to computational cancer biology include new models, theoretical insights into existing models, novel algorithmic techniques and detailed experimental analysis of various datasets. Specifically, we contribute the following.
• Novel algorithmic techniques for denoising array comparative genomic hybridization data. Our algorithmic results are of independent interest and provide approximation techniques for speeding up dynamic programming.
• Based on empirical findings which strongly indicate an inherent geometric structure in cancer genomic data, we introduce a geometric model for finding subtypes of cancer which overcomes difficulties of existing methods such as principal/independent component analysis and separation methods for Gaussians. We provide a computational method which solves the optimization problem efficiently and is robust to outliers.
• We find the necessary and sufficient conditions to reconstruct uniquely an oncogenetic tree, a popular tumor phylogenetic method."

Some of the interesting bits out of the numerous interesting information in this thesis, can be found in how graphs evolve over time.



Page 244, Chapter 13 also features open problems.
Also not directly related, well kinda, the List of Open Problems in Sublinear Algorithms has been updated since the recent 2014 Bertinoro Workshop. Here they are:

Graph, streaming/turnstile, RNA.....

The slides of the Bertinoro workshop are here.

Join the CompressiveSensing subreddit or the Google+ Community and post there !
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.

Saturday, July 05, 2014

Saturday Morning Videos: Paris-Saclay Center for Data Science kick-off meeting

The Paris-Saclay Center for Data Science kick-off meeting occured on Monday at Saclay (south of Paris). Eventually, I am sure the Webcast CC-IN2P3 entity will need to realize that allowing people to embed their videos is a good thing. Without further ado, here are the presentations (clicking on the title of the title will lead to the video).


Intervenant: Patricio Leboeuf (FCS)
Documents: Slides

Intervenant: Balázs Kégl (LAL)
Documents: Slides

Intervenant: Jean-Philippe Vert (Mines ParisTech / Institut Curie)
Documents: Transparents
The rapid technological developments in biology, in particular of DNA sequencing technologies, allow us to collect large amounts of molecular data about the genome of each individual, and opens the possibility to predict drug response or evaluate the risk of various diseases from one's molecular identity. In this talk I will discuss some regularization-based approaches we have developed to estimate complex, high-dimensional predictive models from relatively few samples, in particular in cancer prognosis and toxicogenetics.

Intervenant: Bertrand Thirion (INRIA / Neurospin)
Documents: Slides
Functional brain imaging offers a unique view on brain functional organization, which is broadly characterized by two features: the segregation of brain territories into functionally specialized regions, and the integration of these regions into networks of coherent activity. Among other observation modalities, magnetic resonance imaging yields a spatially resolved, yet noisy view of this organization. In this talk, I will discuss how the use of multiple datasets and machine learning tools can enhance the inference procedures that are necessary to go from data to knowledge on the brain.
Intervenant: Frédéric Schmidt (GEOPS / UPSud)
Documents: Transparents
Remote sensing is the major technique to study planetary environment in order to decipher the structure and evolution of solar system bodies. For a decade, spacecrafts have acquired high-resolution spectra, high-resolution images, hyperspectral images, and multi-angular hyperspectral images. The treatment of raw data to produce high level science results but also the visualization of the large amount of data require innovative tools. Here I review some aspects of data science projects in planetary science, focusing on multi-angular hyperspectral imaging (~500 wavelength), digital terrain model using stereoscopic techniques on high resolution images (~0.5m/pixel), and data visualization.

Intervenant: Prof. Reza ANSARI (LAL-Univ.ParisSud , IN2P3-CNRS)
Documents: Transparents
The Big Bang cosmological model provides a powerful framework to describe the evolution of the Universe. Despite tremendous theoretical and observational progress in the field, profound mysteries such as the nature of dark matter and dark energy remain to be unveiled. After a brief introduction on cosmology, an overview of some of the large projects in astrophysics and cosmology in the next decade will be presented. These projects cover a broad range of the electromagnetic spectrum, from optical surveys (LSST, eBOSS, EUCLID), to future CMB (Cosmic Microwave Background, CORE2) missions and next generation radio interferometers (SKA). Some of the computing challenges faced by these projects will be highlighted, focusing on the LSST (Large Synoptic Survey Telescope) data management and processing case.
Intervenant: Maarten de Rijke (University of Amsterdam)
In the talk I will discuss work on learning to rank for information retrieval, in which the goal is to automatically construct a model that ranks documents in response to a query. In traditional supervised machine learning approaches for the LTR problem one manually selects a set of manually engineered ranking features and then learns the best way of combining them to obtain the most powerful ranking model that those features are capable of producing. In ongoing work on truly autonomous search engines, we are moving evaluation, learning and feature engineering to a weakly supervised paradigm, learning from the implicit feedback that naturally emerges as part of users' interactions with the search engine. I will discuss recent progress in each of these three dimensions: evaluation, learning and feature engineering.

Intervenant: Nicolas Vayatis (ENS Cachan)
Documents: Slides
In every sector of human activity, the pervasiveness of sensors and the accumulation of digital information have raised novel intellectual challenges, dreams and fears. Recently, intensive research in the field of high dimensional statistics, the progress in the description and modeling of networks, and the second life of optimization theory have generated concepts and algorithms that allow to develop inference on complex data and also to think about new perspectives of interactions between experts or scientists of different fields. A major tension when addressing such issues from the viewpoint of applications is the balance between customization and reproducibility and, to my opinion, these two criteria should drive future innovations in the field of machine learning. In the talk, I will illustrate these ideas by going through a few recent achievements arising from interdisciplinary projects in the fields of digital marketing, fluid mechanics, and ethomics.

Intervenant: Juan Pablo Bello (New York University / Telecom ParisTech)

This talk discusses a mix of concepts, problems and techniques at the crossroads of signal processing, machine learning and music. I will start by introducing content-based music information retrieval (MIR) as an important and challenging data science problem. Then, I will discuss recent work done at my lab on a variety of MIR problems such as automatic chord recognition, music structure analysis, cover song identification and instrument recognition. In the process of doing so, I'll review the impact of feature design for specific MIR tasks, suggest that existing feature extraction methods in audio can be re-conceptualized as deep, multi-layer and trainable systems combining affine transforms and subsampling operations, and show a few examples where deep learning matches or outperforms the current state of the art in music and sound classification. Finally, I’ll discuss open challenges and opportunities in the field.

Intervenant: Dr. Tobias Isenberg (Inria)
Since the size and complexity of scientific datasets is growing at a very high rate, people are working on developing techniques to effectively depict and visualize them. However, frequently it is not sufficient to just produce a single static visualization but instead we have to support scientists in discovering aspects about the data that they did not know about it. That means that we have to develop effective interactive visualization tools that support scientists in exploring their data. In my talk I will address the problem of interactively visualizing data that has an inherent mapping to the 3D spatial domain such as MRI scans, physical simulations, or molecular models. Specifically, I use interfaces on large, touch-sensitive displays because they tend to give people the feeling of "being in control of their data." That means we face the problem of providing input on a two-dimensional surface which needs to be mapped to manipulations of the three-dimensional data space. I will talk about FI3D, a technique to navigate in 3D datasets and control 7 degrees of freedom with only one or two fingers being used simultaneously. Next, I will discuss the problem of spatial data selection which is fundamental to further data analysis and also requires to define a 3D selection space with only input on a 2D plane. Finally, I discuss a case study in which we integrated several different interaction techniques into a tool for fluid mechanics experts to explore their data. I will end my talk by pointing out some open problems and research challenges that we are currently facing.

Intervenant: Kyle Cranmer (New York University)
Documents: Transparents
Particle physics poses several unique challenges for data science with multi-petabyte datasets, complex particle detectors, and the search for exceedingly rare signals in the data. The field is characterized by large, international collaborations, which requires a high-level of collaboration. I will give an overview of our data science challenges and discuss the statistical aspects of the recent discovery of the Higgs boson, including the collaborative statistical modeling techniques that are transforming the field. I will identify places where our tools and techniques are quickly evolving or beginning to fail and opportunities for fruitful collaboration with the nascent field of data science.


Intervenant: Akin Kazakci (Mines ParisTech)
Documents: Transparents

Center for Data-Science (CDS) initiatives seem to pop up all around the globe at the moment. Considering the data deluge phenomena, the motivation behind such initiatives may seem trivial. However, a closer look reveals that the purpose, success conditions and managerial principles for CDS initiatives are much less clear. CDSs are neither private companies, nor traditional research entities. What would be a suitable organizational model and philosophy – designed to avoid pitfalls other science-based movements have faced in the history? Beyond the seemingly trivial purpose of being (analytical) service providers, each such initiative needs to build their own strategy for survival, success and long-term impact. They also need to accomplish this feat in a way to differentiate themselves from other initiatives. To this end, the body of knowledge produced by management science in the form of methods, organizational models and best practices can be helpful. This talk will focus on some potential pitfalls and the potential contribution of design theory and innovation management methods for CDS initiatives.

Présidents de session: Mr. Erwan Le Pennec (Ecole Polytechnique), Prof. Michalis Vazirgiannis (LIX Ecole Polytechnqiue)
Documents: slides



Join the CompressiveSensing subreddit or the Google+ Community and post there !
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.

Saturday Morning Video: A simple heart rate monitor, Fireworks filmed with a drone, Thermal Touch - A New Augmented Reality Interface for Wearables






Join the CompressiveSensing subreddit or the Google+ Community and post there !
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.

Friday, July 04, 2014

A Second Inflection Point in Genome Sequencing ? and then ...

It's Friday afternoon, it Hamming's time.

If you read Nuit Blanche (Predicting the Future: The Steamrollers ), you are probably more aware of how the future unfolds. But probably not enough to figure out this upcoming game score.


The good folks at Google predict France will win based on touch by touch data. Anyway, we'll know in a few hours. What about looking into the future farther ahead ?




In genome sequencing, current techniques usually require the chemical cut of the DNA in small pieces to be eventually algorithmically reasembled into a sequence (see the Partial Digest Problem in Reconstruction of Integers from Pairwise Distances). A first inflection point in the democratization of DNA sequencing could be noticed back in 2007-2008 thanks to process parallelization [2]. Another is taking place right before our eyes and is probably going to be more prominent as massive data coming from Nanopore technology [1] unfolds. From Nanopores are here!:
In what appears to be the first example of publicly available, user-generated Oxford Nanopore MinION data, Nick Loman (aka @pathogenomenick) has given us a glimpse into the future.

Let us remember that previously on Nuit Blanche, we could already get our hands on raw data from a similar technology (Quantum Biosystems Provides Raw Data Access to New Sequencing Technology).

But soon enough, getting all this information for every person will mean that we will be able to search for the outliers (rare diseases) first [7] and search for actual drug that can target molecular networks [6] i.e; have more Stephanie Events with algorithms mentioned here on Nuit Blanche.

From [4]

For instance, recently The SwAMP Thing! entry pointed to the possibility of using AMP algorithms to perform group testing, in another entry ( ... There will be a "before" and "after" this paper ...,), other authors have evaluated what population sampling requirements when performing GWAS based on our current solvers capabilities. Again, all techniques and algorithms often featured here...

[5] Applying compressed sensing to genome-wide association studies by Shashaank Vattikuti, James J Lee, Christopher C Chang, Stephen D H Hsu and Carson C Chow

The study of molecular networks has recently moved into the limelight of biomedical research. While it has certainly provided us with plenty of new insights into cellular mechanisms, the challenge now is how to modify or even restructure these networks. This is especially true for human diseases, which can be regarded as manifestations of distorted states of molecular networks. Of the possible interventions for altering networks, the use of drugs is presently the most feasible. In this mini-review, we present and discuss some exemplary approaches of how analysis of molecular interaction networks can contribute to pharmacology (e.g., by identifying new drug targets or prediction of drug side effects), as well as list pointers to relevant resources and software to guide future research. We also outline recent progress in the use of drugs for in vitro reprogramming of cells, which constitutes an example par excellence for altering molecular interaction networks with drugs.

Abstract 
Background
Genome-wide association studies have revealed that rare variants are responsible for a large portion of the heritability of some complex human diseases. This highlights the increasing importance of detecting and screening for rare variants. Although the massively parallel sequencing technologies have greatly reduced the cost of DNA sequencing, the identification of rare variant carriers by large-scale re-sequencing remains prohibitively expensive because of the huge challenge of constructing libraries for thousands of samples. Recently, several studies have reported that techniques from group testing theory and compressed sensing could help identify rare variant carriers in large-scale samples with few pooled sequencing experiments and a dramatically reduced cost.
Results
Based on quantitative group testing, we propose an efficient overlapping pool sequencing strategy that allows the efficient recovery of variant carriers in numerous individuals with much lower costs than conventional methods. We used random k-set pool designs to mix samples, and optimized the design parameters according to an indicative probability. Based on a mathematical model of sequencing depth distribution, an optimal threshold was selected to declare a pool positive or negative. Then, using the quantitative information contained in the sequencing results, we designed a heuristic Bayesian probability decoding algorithm to identify variant carriers. Finally, we conducted in silico experiments to find variant carriers among 200 simulated Escherichia coli strains. With the simulated pools and publicly available Illumina sequencing data, our method correctly identified the variant carriers for 91.5-97.9% variants with the variant frequency ranging from 0.5 to 1.5%.
Conclusions
Using the number of reads, variant carriers could be identified precisely even though samples were randomly selected and pooled. Our method performed better than the published DNA Sudoku design and compressed sequencing, especially in reducing the required data throughput and cost.




Join the CompressiveSensing subreddit or the Google+ Community and post there !
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.

SASIR: LOFAR Sparse Image Reconstruction - implementation -

Interesting, Compressive Sensing is becoming mainstream in Big Science.




LOFAR Sparse Image Reconstruction by H. Garsden, J. N. Girard, J. L. Starck, S. Corbel, C. Tasse, A. Woiselle, J. McKean, A.S. van Amesfoort, J. Anderson, I. M. Avruch, R. Beck, M. J. Bentum, P. Best, F. Breitling, J. Broderick, M. Brüggen, H. R. Butcher, B. Ciardi, F. de Gasperin, E. de Geus, M. de Vos, S. Duscha, J. Eislöffel, D. Engels, H. Falcke,R. A. Fallows, R. Fender, C. Ferrari, W. Frieswijk, M. A. Garrett, J. Grießmeier, A. W. Gunst, T. E. Hassall,G. Heald, M. Hoeft, J. Hörandel, A. van der Horst, E. Juette, A. Karastergiou, V. I. Kondratiev, M. Kramer,M. Kuniyoshi, G. Kuper, G. Mann, S. Markoff, R. McFadden, D. McKay-Bukowski, D. D. Mulcahy, H. Munk,M. J. Norden, E. Orru, H. Paas, M. Pandey-Pommier, V. N. Pandey, G. Pietka, R. Pizzo, A. G. Polatidis, A. Renting, H. Röttgering, A. Rowlinson, et al. (21 additional authors not shown)

Context. The LOFAR Radio Telescope is a giant digital phased array interferometer with multiple antennas gathered in stations placed throughout Europe. As other interferometers, it provides a discrete set of measured Fourier components of the sky brightness. With these samples, recovering the original brightness distribution with aperture synthesis forms an inverse problem that can be solved by different deconvolution and minimization methods. Aims. Recent papers have established a clear link between the discrete nature of radio interferometry measurement and "compressed sensing" theory, which supports sparse recovery methods to reconstruct an image from the measured visibilities. We aimed at the implementation and at the scientific validation of one of these methods. Methods. We evaluated the photometric and resolution performance of the sparse recovery method in the framework of the LOFAR instrument on simulated and real data. Results. We have implemented a sparse recovery method in the standard LOFAR imaging tools, allowing us to compare the reconstructed images from both simulated and real data with images obtained from classical methods such as CLEAN or MS-CLEAN. Conclusions.We show that i) sparse recovery performs as well as CLEAN in recovering the flux of point sources, ii) performs much better on extended objects (the root mean square error is reduced by a factor up to 10), and iii) provides a solution with an effective angular resolution 2-3 times better than the CLEAN map. Applied to a real LOFAR dataset, the sparse recovery has been validated with the correct photometry and realistic recovered structures of Cygnus A, as compared to other methods. Sparse recovery has been implemented as an image recovery method for the LOFAR Radio Telescope and it can be used for other radio interferometers.

The attendant CS code (SASIR) is here.

Join the CompressiveSensing subreddit or the Google+ Community and post there !
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.

Thursday, July 03, 2014

Rows vs Columns for Linear Systems of Equations - Randomized Kaczmarz or Coordinate Descent ?

Further studies on the Kaczmarz algorithm:


This paper is about randomized iterative algorithms for solving a linear system of equations Xβ=y in different settings. Recent interest in the topic was reignited when Strohmer and Vershynin (2009) proved the linear convergence rate of a Randomized Kaczmarz (RK) algorithm that works on the rows of X (data points). Following that, Leventhal and Lewis (2010) proved the linear convergence of a Randomized Coordinate Descent (RCD) algorithm that works on the columns of X (features). The aim of this paper is to simplify our understanding of these two algorithms, establish the direct relationships between them (though RK is often compared to Stochastic Gradient Descent), and examine the algorithmic commonalities or tradeoffs involved with working on rows or columns. We also discuss Kernel Ridge Regression and present a Kaczmarz-style algorithm that works on data points and having the advantage of solving the problem without ever storing or forming the Gram matrix, one of the recognized problems encountered when scaling kernelized methods.


Let us note that version 3 of reference 6 has been out for the past month and a half:


We obtain an improved finite-sample guarantee on the linear convergence of stochastic gradient descent for smooth and strongly convex objectives, improving from a quadratic dependence on the conditioning (L/μ)2(where L is a bound on the smoothness and μ on the strong convexity) to a linear dependence on L/μ. Furthermore, we show how reweighting the sampling distribution (i.e. importance sampling) is necessary in order to further improve convergence, and obtain a linear dependence in the average smoothness, dominating previous results. We also discuss importance sampling for SGD more broadly and show how it can improve convergence also in other scenarios. Our results are based on a connection we make between SGD and the randomized Kaczmarz algorithm, which allows us to transfer ideas between the separate bodies of literature studying each of the two methods. In particular, we recast the randomized Kaczmarz algorithm as an instance of SGD, and apply our results to prove its exponential convergence, but to the solution of a weighted least squares problem rather than the original least squares problem. We then present a modified Kaczmarz algorithm with partially biased sampling which does converge to the original least squares solution with the same exponential convergence rate.

Randomized algorithms that base iteration-level decisions on samples from some pool are ubiquitous in machine learning and optimization. Examples include stochastic gradient descent and randomized coordinate descent. This paper makes progress at theoretically evaluating the difference in performance between sampling with- and without-replacement in such algorithms. Focusing on least means squares optimization, we formulate a noncommutative arithmetic-geometric mean inequality that would prove that the expected convergence rate of without-replacement sampling is faster than that of with-replacement sampling. We demonstrate that this inequality holds for many classes of random matrices and for some pathological examples as well. We provide a deterministic worst-case bound on the gap between the discrepancy between the two sampling models, and explore some of the impediments to proving this inequality in full generality. We detail the consequences of this inequality for stochastic gradient descent and the randomized Kaczmarz algorithm for solving linear systems.

Image Credit: NASA/JPL-Caltech
This image was taken by Rear Hazcam: Left B (RHAZ_LEFT_B) onboard NASA's Mars rover Curiosity on Sol 677 (2014-07-02 16:42:12 UTC).
Full Resolution

Some Comments: Universal Compressed Sensing of Markov Sources, A survey of uncertainty principles and some signal processing applications, Refined support and entropic uncertainty inequalities


Benjamin Ricaud commented:

Hi Igor,
To add some complementary info to your nice post: you may be interested in our study of the uncertainty principle in signal processing (and the references we cite therein). One is a review of some versions of the uncertainty principles:
http://arxiv.org/abs/1211.5914
We explain the relation between l^p-norms and Shannon entropy via the Renyi entropies (already noticed in the paper you cite) and give some other insights. You are right, these quantities give a family of uncertainty principles which are different from the Heisenberg one in that they do not take into account any localization of a function around a particular value (its mean position). Indeed, if you change the labelling of points in the function domain, the norm do not change (l^p norm or entropy will stay the same), but the variance of the function will change.
We also wrote a paper on uncertainty in 2 different frames, a generalization of the ones using 2 different bases:
The bound in the uncertainty inequality depends on the coherence between the two frames. the same coherence which is also involved in compressive sensing. We also propose a slightly different definition for the coherence.
I hope this will help understanding better the uncertainty principle.
We also prepare a study of some uncertainty principles applying to signal defined on graphs for the following weeks! I send you a mail when it is ready.
Thanks for sharing your views on the uncertainty principle.


and he subsequently added
A few additional facts I want to mention on the uncertainty principle, that I was happy to learn.
If we look at the uncertainty principle between the signal and its Fourier transform. The minimizer of the Heisenberg uncertainty principle (the function which makes the uncertainty equal to the bound) is a Gaussian function, and it is the same for the entropic uncertainty principle in infinite dimension (continuous case). In the finite dimensional case, the minimizer of the l^p-norm family of uncertainty principle (in particular l^0-norm) is a « picket fence signal » , i.e. regularly spaced Kronecker deltas. A striking result is that the Fourier transform of a Gaussian is a Gaussian and the (discrete) Fourier transform of a picket fence is a picket fence. Also an important thing to remark is the difference between continuous and discrete domains. One minimizer is well localized around a mean value whereas the other is completely delocalized, with a few peaks of energy but evenly distributed in the signal domain. If you make a function which approximate well a gaussian in the discrete case, you will have a small uncertainty, but the picket fence beats it. In the continuous case, the picket fence can not exists as it is not square integrable.
Continuing with the continuous vs discrete case, it is difficult to find equivalent of the Heisenberg and its variance measure in the discrete case. You have to adapt the definition to each domain (if it is possible). Even in the continuous case, take functions living on a circle: if you choose a point of the circle to be your zero coordinate and you define your domain to be [-pi,pi] with periodic boundary conditions. If you translate a nice well localized function to a point around pi. A part of the function will appear around -pi because we are on a circle. If you compute the standard variance as defined for (-infinity,infinity), it will give a huge value, but on the circle the function is still the same, with the same concentration.
I did not find any general formula for the variance in the discrete case.
Hence, we have to be careful when comparing uncertainty in the continuous and the discrete case.

The uncertainty principle exists between two representations which are the expressions of the function in two different bases of a Hilbert space. But it exists also for two frames (at least in the finite dimensional case), meaning that you have an uncertainty principle each time you use two linear transformations which are invertible (even if you have to use a pseudo-inverse). That means, the uncertainty principle is related also to information. If you loose some information by looking at your function in some representation you don’t have any uncertainty principle (the bound is zero).
Thomas Arildsen also added:

Rényi's information dimension is also used in this very recent paper on a universal approach to compressed sensing: http://arxiv.org/abs/1406.7807 which also cites [2].

To summarize, we have:


Universal Compressed Sensing of Markov Sources by Shirin Jalali, H. Vincent Poor

The main promise of compressed sensing is accurate recovery of high-dimensional structured signals from an underdetermined set of randomized linear projections. Several types of structure such as sparsity and low-rankness have already been explored in the literature. For each type of structure, a recovery algorithm has been developed based on the properties of signals with that type of structure. Such algorithms recover any signal complying with the assumed model from its sufficient number of linear projections. However, in many practical situations the underlying structure of the signal is not known, or is only partially known. Moreover, it is desirable to have recovery algorithms that can be applied to signals with different types of structure.
In this paper, the problem of developing "universal" algorithms for compressed sensing of stochastic processes is studied. First, Renyi's notion of information dimension is generalized to analog stationary processes. This provides a measure of complexity for such processes and is connected to the number of measurements required for their accurate recovery. Then a minimum entropy pursuit (MEP) algorithm is proposed, and it is proven that it can reliably and robustly recover any Markov process of any order from sufficient number of randomized linear measurements, without having any prior information about the distribution of the process. An implementable version of MEP with the same performance guarantees is also provided.



The goal of this paper is to review the main trends in the domain of uncertainty principles and localization, emphasize their mutual connections and investigate practical consequences. The discussion is strongly oriented towards, and motivated by signal processing problems, from which significant advances have been made recently. Relations with sparse approximation and coding problems are emphasized.

Generalized versions of the entropic (Hirschman-Beckner) and support (Elad-Bruckstein) uncertainty principle are presented for frames representations. Moreover, a sharpened version of the support inequality has been obtained by introducing a generalization of the coherence. In the finite dimensional case and under certain conditions, minimizers of this inequalities are given as constant functions on their support. In addition, ℓp-norms inequalities are introduced as byproducts of the entropic inequalities.
Thanks Thomas and Benjamin !

Join the CompressiveSensing subreddit or the Google+ Community and post there !
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.

Wednesday, July 02, 2014

Entropic Uncertainty Relations in Quantum Physics / On asymptotic structure in compressed sensing

I am having a small discussion with Greg Howland one of the authors of yesterday's paper ( Simultaneous Measurement of Complementary Observables with Compressive Sensing )

....[Everything that is about to be said here is based on my own current limited understanding]....

One item he points me to is the following paper on the use of Shannon entropy in Quantum Mechanics. To make a long story short, some of the readers might recognize issues in this example ( see figure below ) and its attendant conclusion ( see below):


...Therefore, σx tends to infinity with N → ∞, while common sense predicts that the uncertainty should remain finite. So, why the uncertainty measured by the standard deviation grows with N? It is simply the effect of a growing distance between the two regions. Standard deviation is based on the second moment which is very sensitive to the values of an observable lying far away from the average. This is the first flaw of the standard deviation...

For short, the Heisenberg principle uses standard deviations and it is not helpful because it is based on the L2 norm and is therefore sensitive to outliers. As an outsider, I just hope that all the Quantum Mechanics "paradoxes" are of the same nature. Hence according to the paper, the Heisenberg principle can be recasted using a Shannon entropy argument where it makes more common sense. But while I am reading the rest of the paper, I get the sense that if you use those Shannon/Renyi entropies in Quantum Mechanics, we directly go back to traditional issues that have been studied in compressive sensing:

The physical interpretation of the inequality (79) is similar to the Heisenberg uncertainty relation. Two observables A and B, characterized by the bases |aii and |bji cannot have simultaneously sharp values since they do not have common basis vectors. Moreover, every state described by a basis vector has the sum of the uncertainties equal to the same value 1/√N. The observables A and B are finite dimensional analogs of canonically conjugate physical quantities.
The observables or the measurements cannot have similarly sharp values (the main argument of the Heisenberg principle) is just a statement on basis incoherence seen in compressive sensing. If you recall this Ghost imaging experiment back in 2009, you'll notice that one could already reconstruct the image using equation 2 of this paper without any assumption on the sparsity of the scene. In other words, while the measurements are sharp for one variable (raster scan) and "diffuse" for the other (random group or CS scan), one can already reconstruct a sharp/sparse variable from the "diffuse" measurement thanks to the randomness of the filter alone.

What does Compressive Sensing bring to the table ? 

Well, the additional assumption of sparsity on what is being measured, allows fewer "diffuse" measurements to be deconvoluted back to the sparse (sharp) value of the variable of interest. This was well stated by Emmanuel Candes and Terry Tao back in 2008 in a presentation entitled The uniform uncertainty principle and compressed sensing:

(Note that one normally needs all p Fourier coefficients in order to recover a general signal; the point is that sparse signals have a much lower information entropy and thus are easier to recover than general signals.)
(For more details, please check Terry Tao's 2007 blog entry and attendant comments)

Let us note that within this description, incoherence in Quantum Mechanics and incoherence in Compressive Sensing have identical meanings. Let us also note that the assumption of sparsity is strong in that one does not require random filters to reconstruct the variable of interest (there are deterministic measurement ensembles or some other structurally random measurement ensembles that can be used in compressive sensing). Random filters are, however, required to reconstruct a variable of interest without the sparsity assumption.

Of note, the field has evolved ever since the earliest papers on compressive sensing and there has been an on-going discussion and progress on this issue of incoherence. As a matter of fact, we seem to converge toward a better understanding of it [1] while other approaches that do not rely on incoherence properties are also making strides by showing how sharp phase transitions produce a limit on actual hardware/measurement systems. Let us hope that this on-going debate can eventually include the Quantum Mechanics and Quantum Optics folks. For instance, in QM, the uncertainty principle is also equivalent to the fact that two operators do not commute. Can we design specific measurement matrices based on the diverse findings in compressive sensing and group testing to get a good evaluation of both variables ? How about using sharp phase transitions in order to evaluate the limits of states that can be probed with compressive sensing solvers. The Renyi entropy proposed in the paper has been looked into in compressive sensing [2] and attendant solvers have been found that are optimal for that [3] (AMP solvers). If this interpretation holds, then suddenly QM could become much clearer to a larger swath of the research community. Without further ado, here is the paper:

Entropic Uncertainty Relations in Quantum Physics by Iwo Bialynicki-Birula, Lukasz Rudnicki

Uncertainty relations have become the trademark of quantum theory since they were formulated by Bohr and Heisenberg. This review covers various generalizations and extensions of the uncertainty relations in quantum theory that involve the R\'enyi and the Shannon entropies. The advantages of these entropic uncertainty relations are pointed out and their more direct connection to the observed phenomena is emphasized. Several remaining open problems are mentioned.



[1] On asymptotic structure in compressed sensing by Bogdan Roman, Anders Hansen, Ben Adcock

This paper demonstrates how new principles of compressed sensing, namely asymptotic incoherence, asymptotic sparsity and multilevel sampling, can be utilised to better understand underlying phenomena in practical compressed sensing and improve results in real-world applications. The contribution of the paper is fourfold:
First, it explains how the sampling strategy depends not only on the signal sparsity but also on its structure, and shows how to design effective sampling strategies utilising this.
Second, it demonstrates that the optimal sampling strategy and the efficiency of compressed sensing also depends on the resolution of the problem, and shows how this phenomenon markedly affects compressed sensing results and how to exploit it.
Third, as the new framework also fits analog (infinite dimensional) models that govern many inverse problems in practice, the paper describes how it can be used to yield substantial improvements.
Fourth, by using multilevel sampling, which exploits the structure of the signal, the paper explains how one can outperform random Gaussian/Bernoulli sampling even when the classical l1 recovery algorithm is replaced by modified algorithms which aim to exploit structure such as model based or Bayesian compressed sensing or approximate message passaging. This final observation raises the question whether universality is desirable even when such matrices are applicable.
Examples of practical applications investigated in this paper include Magnetic Resonance Imaging (MRI), Electron Microscopy (EM), Compressive Imaging (CI) and Fluorescence Microscopy (FM). For the latter, a new compressed sensing approach is also presented.


Join the CompressiveSensing subreddit or the Google+ Community and post there !
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.

Stable, Robust and Super Fast Reconstruction of Tensors Using Multi-Way Projections - implementation -




We introduce an analytical reconstruction formula that allows one to recover an Nth-order data tensor from a reduced set of multi-way compressive measurements by exploiting its low multilinear-rank structure. Moreover, it is shown that, an interesting property of multi-way measurements allows us to build the reconstruction based on compressive linear measurements of fibers taken only in two selected modes, independently of the tensor order N. In addition, it is proved that, in the matrix and 3rd-order tensor cases, the proposed reconstruction X−−τ is stable in the sense that the approximation error is comparable to the one provided by the best low-multilinear-rank approximation, i.e ∥X−−−X−−^τ∥≤K∥X−−−X−−0∥, where τ is a threshold parameter that controls the approximation error, K is a constant and X−−0 is the best low multilinear-rank approximation of X−−. Through the analysis of the upper bound of the approximation error, we find that, in the 2D case, an optimal value for the threshold parameter τ=τ0≠0 exists. On the other hand, in the 3D case, the optimal value for this parameter is τ=0 which means that this parameter does not need to be tuned in order to obtain good reconstructions. We also present extensive simulation results indicating the stability and robustness of the method when it is applied to real-world 2D and 3D signals. A comparison with state of the arts Compressed Sensing (CS) methods specialized for multidimensional signals is also included. A very attractive characteristic of the proposed method is that it provides a direct computation, i.e. it is non-iterative unlike all existing CS algorithms, thus providing super fast computations, even for large datasets.
I recently asked the authors whether they would make available their important implementations. Cesar Caiafa kindly responded with:

Dear Igor,
I just want to let you know that our preprint is now updated in ArXiv (http://arxiv.org/abs/1406.3295).
Now it includes a link to our Matlab code available at
http://web.fi.uba.ar/~ccaiafa/Cesar/Low-Rank-Tensor-CS.html
Thank you very much.
Best Regards


Cesar
Thank you Cesar !



We review and introduce new representations of tensor train decompositions for large-scale vectors, matrices, or low-order tensors. We provide extended definitions of mathematical multilinear operations such as Kronecker, Hadamard, and contracted products, with their properties for tensor calculus. Then we introduce an effective low-rank tensor approximation technique called the tensor train (TT) format with a number of mathematical and graphical representations. We also provide a brief review of mathematical properties of the TT format as a low-rank approximation technique. With the aim of breaking the curse-of-dimensionality in large-scale numerical analysis, we describe basic operations on large-scale vectors and matrices in TT format. The suggested representations can be used for describing numerical methods based on the TT format for solving large-scale optimization problems such as the system of linear equations and eigenvalue problems.


Join the CompressiveSensing subreddit or the Google+ Community and post there !
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.

Tuesday, July 01, 2014

Simultaneous Measurement of Complementary Observables with Compressive Sensing

Dick Gordon and several other readers mentioned the following paper to me. It aims at doing something that was thought to be impossible: i.e. measuring both position and momentum of a system at the same time. From afar, the Heisenberg principle says it cannot be done but as the authors mention:

...We strongly emphasize that our technique does not violate the uncertainty principle; at no point does a single detection event give precise information about both position and momentum. Instead, each detection event gives some information about both domains. Our approach economizes the use of this information... 

So if I understand correctly, the Heisenberg principle is somehow not applicable here because:
  • the authors perform two detections (not one) and, in my view the next argument seems the most important assumption,
  • the authors use the additional information about the sparsity of the event so that one can use the random projections of the masks. Had the event not been sparse, or in other words, had they had to detect many several positions and momentums, position measurements using random masks would not yield any position information because that information would not be sparse (and therefore algorithmic position reconstruction would fail).

Limits on these systems can therefore directly use current known limits of the reconstruction solvers (also known as the Donoho-Tanner sharp phase transition) and this with Hadamard matrices or even more economical measurement matrices. Suddenly, 10 years later, sharp phase transitions from combinatorics yield another interesting twist....If this interpretation holds, Quantum Mechanics and its many paradoxes might become less obscure. Without further ado (thanks Doug!), here is the paper:




The more information a measurement provides about a quantum system’s position statistics, the less information a subsequent measurement can provide about the system’s momentum statistics. This information trade-off is embodied in the entropic formulation of the uncertainty principle. Traditionally, uncertainly relations correspond to resolution limits; increasing a detector’s position sensitivity decreases its momentum sensitivity and vice versa. However, this is not required in general; for example, position information can instead be extracted at the cost of noise in momentum. Using random, partial projections in position followed by strong measurements in momentum, we efficiently determine the transverse-position and transverse-momentum distributions of an unknown optical field with a single set of measurements. The momentum distribution is directly imaged, while the position distribution is recovered using compressive sensing. At no point do we violate uncertainty relations; rather, we economize the use of information we obtain.



Join the CompressiveSensing subreddit or the Google+ Community and post there !
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.

Printfriendly