Data-Selective Learning for Sparse System Identification

Share

ABSTRACT
This talk addresses fundamental as well as some advanced concepts that are important to understand the principles of data-selective algorithms in sparse system identification. This type of algorithms utilize environmental data for updating system parameters only when they bring new information. As a result, the obtained parameter estimations become more accurate without sacrificing the learning speed. The data-selective algorithms are particularly suitable for applications where computational resources are limited and/or power saving is a requirement.
Two adaptive filtering algorithms that combine a sparsity-promoting scheme with a data-selection mechanism are presented. Sparsity is promoted via approximations to the l0 norm in order to increase convergence speed of the algorithms when dealing with sparse/compressible signals. These approximations circumvent some difficulties of working with the l0 norm, thus allowing the development of online algorithms. Data selection is implemented based on set-membership filtering, which yields robustness against noise and reduced computational burden. These algorithms are analyzed in order to set properly their parameters to guarantee stability. In addition, we characterize their updating processes from a geometrical viewpoint. Simulation results show that the new algorithms outperform the state-of-the-art algorithms designed to exploit sparsity.



  Date and Time

  Location

  Hosts

  Registration



  • Montreal, Rio de Janeiro
  • Canada
  • Building: Concordia University, EV Building
  • Room Number: EV 003.309

Staticmap?size=250x200&sensor=false&zoom=14&markers=45.4697382%2c 73


  Speakers

PAULO S. R. DINIZ of Federal University of Rio de Janeiro

Topic:

Data-Selective Learning for Sparse System Identification

ABSTRACT
This talk addresses fundamental as well as some advanced concepts that are important to understand the principles of data-selective algorithms in sparse system identification. This type of algorithms utilize environmental data for updating system parameters only when they bring new information. As a result, the obtained parameter estimations become more accurate without sacrificing the learning speed. The data-selective algorithms are particularly suitable for applications where computational resources are limited and/or power saving is a requirement.
Two adaptive filtering algorithms that combine a sparsity-promoting scheme with a data-selection mechanism are presented. Sparsity is promoted via approximations to the l0 norm in order to increase convergence speed of the algorithms when dealing with sparse/compressible signals. These approximations circumvent some difficulties of working with the l0 norm, thus allowing the development of online algorithms. Data selection is implemented based on set-membership filtering, which yields robustness against noise and reduced computational burden. These algorithms are analyzed in order to set properly their parameters to guarantee stability. In addition, we characterize their updating processes from a geometrical viewpoint. Simulation results show that the new algorithms outperform the state-of-the-art algorithms designed to exploit sparsity.

Biography:

Paulo S. R. Diniz was born in Niterói, Brazil. He received his Electronics Eng. degree (Cum Laude) from the Federal University of Rio de Janeiro (UFRJ) in 1978, his M.Sc. degree from COPPE/UFRJ n 1981, and his Ph.D. from Concordia University, Montreal, P.Q., Canada, in 1984, all in electrical engineering. He wrote the books ADAPTIVE FILTERING: Algorithms Practical Implementation, Springer, Fourth Edition 2013, and DIGITAL SIGNAL PROCESSING: System Analysis and Design, Cambridge University Press, Cambridge, UK, Second Edition 2010 (with E. A. B. da Silva and S. L. Netto) and the monograph BLOCK TRANSCEIVERS: OFDM and Beyond, Morgan & Claypool, New York, NY, 2012 (W. A. Martins, and M. V. S. Lima). He is a Fellow of IEEE, and a Fellow of EURASIP. He also holds some best-paper awards from conferences and from an IEEE journal. In 2014, he received the Charles A. Desoer Technical Achievements Award from the IEEE Circuits and Systems Society.

Email:

Address:Brazil