Theoretical Limitation for Parallelizing KDD based Algorithms for Data Mining and Proposal of an Alternative Approach

C. Rodríguez Lucatero (Mexico)

Keywords

Computational Complexity, PAC-learning, parallel computing, data mining.

Abstract

In this article we try to study why the parallel versions of KDD based algorithms normally used in Data mining don’t have a good performance and we give some formal arguments for it. Additionally we propose an alternative approach for doing KDD that makes a trade-off between performance and precision using parallel versions of PAC-learning algorithms for learning PAC-learnable concepts (concepts expressed in k-FNC[2][6] and monotonic k-DNF [14], simple decision lists [12][6], equivalence query simulation using less examples [13] , logic recursive programs [15], concepts with finite Vapnik-Chervonenkis dimension [6] ).

Important Links:

• DOI:
• From Proceeding (361) Information and Knowledge Sharing - 2002

Go Back