libsvm_cbe: LIBSVM for the CELL processor
=========================================

Copyright (c) 2000-2009 Chih-Chung Chang and Chih-Jen Lin
Copyright (c) 2009-2011 Moreno Marzolla <marzolla(at)cs.unibo.it>

This software is a modified version of LIBSVM which takes advantage of
the CELL processor. LIBSVM is written by Chih-Chung Chang and Chih-Jen
Lin, and implements training and classification algorithms for Support
Vector Machines. The original version of LIBSVM can be downloaded from
http://www.csie.ntu.edu.tw/~cjlin/libsvm/

libsvm_cbe was written by Moreno Marzolla, based on LIBSVM version
2.89. libsvm_cbe takes advantage of CELL Synergistic Processing
Elements (SPEs) to offload the most computationally intensive task,
that is the computation of the kernel function during the training
step.

libsvm_cbe has been compiled and tested on a Sony PlayStation(R)3
console running Yellow Dog Linux version 5.0; it is not necessary to
install IBM's Cell SDK, as libsvm_cbe only uses tools and libraries
which are already provided by the Yellow Dog Linux
distribution. libsvm_cbe has also been succesfully cross-compiled on a
Linux machine running Ubuntu 9.04/x86.

This software has been described in the following paper:

  M. Marzolla, Fast Training of Support Vector Machines on the Cell
  Processor, Neurocomputing, Volume 14, Issue 17, October 2011,
  pp. 3700—3707, ISSN 0925-2312. DOI 10.1016/j.neucom.2011.04.011

Additional information are available here:

  http://www.moreno.marzolla.name/software/svmcell/
  http://www.cs.unibo.it/pub/marzolla/svmcell/