Tracking time-varying sparse signals is a recent problem with widespread applications. Techniques derived from compressed sensing, Lasso, and Kalman filtering have been proposed in the literature, which mainly present two drawbacks: the prior knowledge of specific evolution models and the lack of theoretical guarantees. In this work, we propose a new perspective on the problem, based on the theory on online convex optimization, which has been developed in the machine learning community. We exploit a strongly convex model, and we develop online algorithms, for which we are able to provide a dynamic regret analysis. A few simulations that support the theoretical results are finally presented.

Online convex optimization meets sparsity / Fosson, Sophie; Javier, Matamoros; Maria, Gregori; Magli, Enrico. - ELETTRONICO. - (2017). (Intervento presentato al convegno Signal Processing with Adaptive Sparse Structured Representations (SPARS) 2017 tenutosi a Lisbona (PT) nel June 5-8, 2017).

Online convex optimization meets sparsity

FOSSON, SOPHIE;MAGLI, ENRICO
2017

Abstract

Tracking time-varying sparse signals is a recent problem with widespread applications. Techniques derived from compressed sensing, Lasso, and Kalman filtering have been proposed in the literature, which mainly present two drawbacks: the prior knowledge of specific evolution models and the lack of theoretical guarantees. In this work, we propose a new perspective on the problem, based on the theory on online convex optimization, which has been developed in the machine learning community. We exploit a strongly convex model, and we develop online algorithms, for which we are able to provide a dynamic regret analysis. A few simulations that support the theoretical results are finally presented.
File in questo prodotto:
File Dimensione Formato  
SPARS2017_Paper_62.pdf

accesso aperto

Tipologia: 2. Post-print / Author's Accepted Manuscript
Licenza: Pubblico - Tutti i diritti riservati
Dimensione 401.69 kB
Formato Adobe PDF
401.69 kB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11583/2673368
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo