Parallelization of Fast 𝓁1-Minimization for Face Recognition

Wagner, Andrew; Shia, Victor; Yang, Allen; Murphy, Mark; Ma, Yi

Permalink

https://hdl.handle.net/2142/74357 Copy

Description

Title

Parallelization of Fast 𝓁1-Minimization for Face Recognition

Author(s)

• Wagner, Andrew
• Shia, Victor
• Yang, Allen
• Murphy, Mark
• Ma, Yi

Issue Date

2011-06

Keyword(s)

Face recognition

Date of Ingest

• 2015-04-06T20:54:51Z
• 2017-07-14T23:10:36Z

Abstract

Recently a family of promising face recognition algorithms based on sparse representation and `1-minimization (`1-min) have been developed. These algorithms have not yet seen commercial application, largely due to higher computational cost compared to other traditional algorithms. This paper studies techniques for leveraging the massive parallelism available in GPU and CPU hardware to accelerate `1-min based on augmented Lagrangian method (ALM) solvers. For very large problems, the GPU is faster due to higher memory bandwidth, while for problems that fit in the larger CPU L3 cache, the CPU is faster. On both architectures, the proposed implementations significantly outperform naive library-based implementations, as well as previous systems. The source code of the algorithms will be made available for peer evaluation.

Publisher

Coordinated Science Laboratory, University of Illinois at Urbana-Champaign

Series/Report Name or Number

Coordinated Science Laboratory Report no. UILU-ENG-11-2204, DC-251

Type of Resource

text

Genre of Resource

Report (Grant or Annual)

Language

English

Permalink

http://hdl.handle.net/2142/74357

Sponsor(s)/Grant Number(s)

ONR / N00014-09-1-0230

Owning Collections