Building Multiclass Nonlinear Classifiers with GPUs
CSAIL, MIT, Cambridge, MA 02139
Big Learning: Advances in Algorithms and Data Management, 2013
@article{arnaldo2013building,
title={Building Multiclass Nonlinear Classifiers with GPUs},
author={Arnaldo, Ignacio and Veeramachaneni, Kalyan and O’Reilly, Una-May},
year={2013}
}
The adoption of multiclass classification strategies that train independent binary classifiers becomes challenging when the goal is to retrieve nonlinear models from large datasets and the process requires several passes through the data. In such scenario, the combined use of a search and score algorithm and GPUs allows to obtain binary classifiers in a reduced time. We demonstrate our approach by training a ten class classifier over more than 400K exemplars following the exhaustive Error Correcting Output Code strategy that decomposes into 511 binary problems.
December 25, 2013 by hgpu
