|
Title:
|
AN EVALUATION OF META LEARNING AND DISTRIBUTION STRATEGIES IN DISTRIBUTED MACHINE LEARNING |
|
Author(s):
|
Andreas D. Lattner, Alexander Grimme, Ingo J. Timm |
|
ISBN:
|
978-972-8939-23-6 |
|
Editors:
|
António Palma dos Reis and Ajith P. Abraham |
|
Year:
|
2010 |
|
Edition:
|
Single |
|
Keywords:
|
Distributed Machine Learning, Meta Learning, Distribution Strategies |
|
Type:
|
Full Paper |
|
First Page:
|
69 |
|
Last Page:
|
76 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
With technical progress in the past decades with multi-core computing, networking, broad availability of computing resources as well as the possibility to store and process huge amounts of data the desire of taking advantage of this situation emerges. In our work, we introduce a distributed learning approach with meta learning and compare different partitioning and majority voting strategies. We focus on strategies that allow for parallel learning and compare different combinations of distribution, meta learning, and voting strategies in an experimental evaluation on two test sets. The results show clear improvement in run time in distributed learning while the results on meta learning exhibit an advantage only in one of the two data set |
|
|
|
|
|
|
