Recovery of Clustered Sparse Signals from Compressive Measurements Volkan Cevher [email protected] Richard BaraniukChinmay Hegde Piotr Indyk.

Recovery of Clustered Sparse Signals from Compressive Measurements

Volkan [email protected]

Richard BaraniukChinmay HegdePiotr Indyk

The Digital Universe

• Size: ~300 billion gigabytes generated in 2007

digital bits > stars in the universegrowing by a factor of 10 every 5 years > Avogadro’s number (6.02x1023) in 15 years

• Growth fueled by multimedia / multisensor data

audio, images, video, surveillance cameras, sensor nets, …

• In 2007 digital data generated > total storage

by 2011, ½ of digital universe will have no home

[Source: IDC Whitepaper “The Diverse and Exploding Digital Universe” March 2008]

Challenges

Acquisition

Compression

Processing

Approaches

• Do nothing / Ignore

be content with where we are…

– generalizes well

– robust

Approaches

• Finite Rate of Innovation

Sketching / Streaming

Compressive Sensing

[Vetterli, Marziliano, Blu; Blu, Dragotti, Vetterli, Marziliano, Coulot; Gilbert, Indyk, Strauss, Cormode, Muthukrishnan; Donoho; Candes, Romberg, Tao; Candes, Tao]

Approaches

• Finite Rate of Innovation

Sketching / Streaming

Compressive Sensing

[Vetterli, Marziliano, Blu; Blu, Dragotti, Vetterli, Marziliano, Coulot; Gilbert, Indyk, Strauss, Cormode, Muthukrishnan; Donoho; Candes, Romberg, Tao; Candes, Tao]

PARSITY

Agenda

• A short review of compressive sensing

• Beyond sparse models

– Potential gains via structured sparsity

• Block-sparse model

• (K,C)-sparse model

• Conclusions

Compressive Sensing 101

• Goal: Recover a sparse orcompressible signal from measurements

• Problem: Randomprojection not full rank

• Solution: Exploit the sparsity/compressibilitygeometry of acquired signal

• Goal: Recover a sparse orcompressible signal from measurements

• Problem: Randomprojection not full rankbut satisfies Restricted Isometry Property (RIP)

• Solution: Exploit the sparsity/compressibility geometry of acquired signal

– iid Gaussian– iid Bernoulli– …

Compressive Sensing 101

• Goal: Recover a sparse orcompressible signal from measurements

• Problem: Randomprojection not full rank

• Solution: Exploit the modelgeometry of acquired signal

Compressive Sensing 101

• Sparse signal: only K out of N coordinates nonzero

– model: union of K-dimensional subspacesaligned w/ coordinate axes

Basic Signal Models

sorted index

• Sparse signal: only K out of N coordinates nonzero

– model: union of K-dimensional subspaces

• Compressible signal: sorted coordinates decay rapidly to zero

well-approximated by a K-sparse signal(simply by thresholding)

sorted index

Basic Signal Models

power-lawdecay

Recovery Algorithms

• Goal:given

recover

• and convex optimization formulations– basis pursuit, Dantzig selector, Lasso, …

• Greedy algorithms– iterative thresholding (IT),

compressive sensing matching pursuit (CoSaMP)subspace pursuit (SP)

– at their core: iterative sparse approximation

Performance of Recovery

• Using methods, IT, CoSaMP, SP

• Sparse signals

– noise-free measurements: exact recovery – noisy measurements: stable recovery

• Compressible signals

– recovery as good as K-sparse approximation

CS recoveryerror

signal K-termapprox error

noise

Sparse Signals

wavelets:natural images

Gabor atoms:chirps/tones

pixels:background subtracted

images

Sparsity as a Model

• Sparse/compressible signal model captures simplistic primary structure

sparse image

• Sparse/compressible signal model captures simplistic primary structure

• Modern compression/processing algorithms capture richer secondary coefficient structure

Beyond Sparse Models

wavelets:natural images

Gabor atoms:chirps/tones

pixels:background subtracted

images

Sparse Signals

• Defn: K-sparse signals comprise a particular set of K-dim canonical subspaces

Model-Sparse Signals

• Defn: A K-sparse signal model comprises a particular (reduced) set of K-dim canonical subspaces

Model-Sparse Signals

• Defn: A K-sparse signal model comprises a particular (reduced) set of K-dim canonical subspaces

• Structured subspaces

<> fewer subspaces

<> relaxed RIP

<> fewer measurements

[Blumensath and Davies]

Model-Sparse Signals

• Defn: A K-sparse signal model comprises a particular (reduced) set of K-dim canonical subspaces

• Structured subspaces

<> increased signal discrimination

<> improved recovery perf.

<> faster recovery

Block Sparse Signals

• Motivation

– sensor networks

• Signal model

intra-sensor: sparsity inter-sensor: common sparse supports

– union of subspaces when signals are concatenated

Block-Sparsity

sensors

[Tropp, Eldar, Mishali; Stojnic, Parvaresh, Hassibi; Baraniuk, VC, Duarte, Hegde]

…

*Individual block sizes may vary…

• Sampling bound (B = # of blocks)

• Problem specific solutions

– Mixed -norm solutions

– Greedy solutions: simultaneous orthogonal matching pursuit [Tropp]

• Model-based recovery framework: -norm [Baraniuk, VC, Duarte, Hegde]

Block-Sparsity Recovery

within block

[Eldar, Mishali (provable); Stojnic, Parvaresh, Hassibi]

• Iterative Thresholding

Standard CS Recovery

[Nowak, Figueiredo; Kingsbury, Reeves; Daubechies, Defrise, De Mol; Blumensath, Davies; …]

update signal estimate

prune signal estimate(best K-term approx)

update residual

• Iterative Model Thresholding

Model-based CS Recovery

[VC, Duarte, Hegde, Baraniuk; Baraniuk, VC, Duarte, Hegde]

update signal estimate

prune signal estimate(best K-term model approx)

update residual

• Provable guarantees

Model-based CS Recovery

update signal estimate

prune signal estimate(best K-term model approx)

update residual

[Baraniuk, VC, Duarte, Hegde]

• Iterative Block Thresholding

Block-Sparse CS Recovery

within blocksort

+ threshold

Block-Sparse Signal

target CoSaMP (MSE = 0.723)

block-sparse model recovery (MSE=0.015) Blocks are pre-specified.

Clustered Sparse Signals:(K,C)-sparsesignal model

Clustered Sparsity

• (K,C) sparse signals (1-D)

– K-sparse within at most C clusters

• Stable recovery

• Recovery:

– w/ model-based framework

– model approximation via dynamic programming(recursive / bottom up)

Example

Simulation via Block-Sparsity

• Clustered sparse <> approximable by block-sparse

*If we are willing to pay 3 × sampling penalty

• Proof by (adversarial) construction

…

Clustered Sparsity in 2D

target Ising-modelrecovery

CoSaMPrecovery

LP (FPC)recovery

• Model clustering of significant pixels in space domain using graphical model (MRF)

• Ising model approximation via graph cuts[VC, Duarte, Hedge, Baraniuk]

Conclusions

• Why CS works: stable embedding for signals with special geometric structure

• Sparse signals >> model-sparse signals

• Greedy model-based signal recovery algorithms

upshot: provably fewer measurementsmore stable recovery

new model: clustered sparsity

• Compressible signals >> model-compressible signals

Volkan Cevher / [email protected]