PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Quasi-Random Resamplings, with aplications to rule Extraction, Cross-Validation and (su-)bagging
Olivier Teytaud, Justin Bedo, Stéphane Lallich and Elie Prudhomme
In: IIIA'06, 2006, Helsinki.

Abstract

Resampling (typically, but not necessarily, bootstrapping) is a well-known stochastic technique for improving estimates in particular for small samples. It is known very efficient in many cases. Its drawback is that resampling leads to a compromise computational cost / stability through the number of resamplings. The computational cost is due to the study of multiple randomly drawn resam- ples. Intuitively, we want some more properly distributed resamples to improve the stability of resampling-based algorithms. Quasi-random numbers are a well- known technique for improving the convergence rate of data-based estimates. We here consider quasi-random version of resamplings. We apply this technique to BSFD, a data-mining algorithm for simultaneous-hypothesis-testing, to cross- validation, and to (su-)bagging, an ensemble method for learning. We present quasi-random numbers in section 2. We present bootstrap and a quasi-random version of bootstrap-sampling in section 3. We present experimental results in section 4.

PDF - Requires Adobe Acrobat Reader or other PDF viewer.
EPrint Type:Conference or Workshop Item (Oral)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Learning/Statistics & Optimisation
ID Code:3192
Deposited By:Olivier Teytaud
Deposited On:20 January 2008