TY - GEN
T1 - Sparse grid classifiers as base learners for AdaBoost
AU - Heinecke, Alexander
AU - Peherstorfer, Benjamin
AU - Pflüger, Dirk
AU - Song, Zhongwen
PY - 2012
Y1 - 2012
N2 - We consider a classification method based on sparse grids which scales only linearly in the number of data points and is thus well-suited for huge amounts of data. In order to obtain competitive results, such sparse grid classifiers are usually enhanced by locally refining the underlying regular sparse grid. However, in order to parallelize the corresponding adaptive algorithms a thorough knowledge of the hardware is necessary. Instead of improving the performance by refining the sparse grid, we construct a team of classifiers relying just on regular sparse grids and employ them as base learners within AdaBoost. Our examples with synthetic and real-world datasets show that we can achieve similar or better results than with locally refined sparse grids or libSVM, with respect to both runtime and accuracy.
AB - We consider a classification method based on sparse grids which scales only linearly in the number of data points and is thus well-suited for huge amounts of data. In order to obtain competitive results, such sparse grid classifiers are usually enhanced by locally refining the underlying regular sparse grid. However, in order to parallelize the corresponding adaptive algorithms a thorough knowledge of the hardware is necessary. Instead of improving the performance by refining the sparse grid, we construct a team of classifiers relying just on regular sparse grids and employ them as base learners within AdaBoost. Our examples with synthetic and real-world datasets show that we can achieve similar or better results than with locally refined sparse grids or libSVM, with respect to both runtime and accuracy.
KW - AdaBoost
KW - classification
KW - data mining
KW - parallelization
KW - shared memory platforms
UR - http://www.scopus.com/inward/record.url?scp=84866978099&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84866978099&partnerID=8YFLogxK
U2 - 10.1109/HPCSim.2012.6266906
DO - 10.1109/HPCSim.2012.6266906
M3 - Conference contribution
AN - SCOPUS:84866978099
SN - 9781467323598
T3 - Proceedings of the 2012 International Conference on High Performance Computing and Simulation, HPCS 2012
SP - 161
EP - 166
BT - Proceedings of the 2012 International Conference on High Performance Computing and Simulation, HPCS 2012
T2 - 2012 10th Annual International Conference on High Performance Computing and Simulation, HPCS 2012
Y2 - 2 July 2012 through 6 July 2012
ER -