Optimisation de plans d'expérience par méthodes particulaires

Abstract : We propose a new stochastic algorithm for Bayesian optimal design in nonlinear and high dimensional models. Like in the recent work of Peter Müller, we turn the optimization problem into a matter of Monte Carlo Markov chains simulations to explore the expected utility surface. The optimal design is then the mode of this surface seen as a probability distribution. Our algorithm mixes a "particles" method to efficiently explore high dimensional multimodal surfaces, with simulated annealing to concentrate the samples near the modes. We test it in a multiple change-point problem for time series data.
Document type :
Conference papers
Complete list of metadatas

https://hal-ineris.archives-ouvertes.fr/ineris-00972424
Contributor : Gestionnaire Civs <>
Submitted on : Thursday, April 3, 2014 - 4:28:31 PM
Last modification on : Thursday, January 11, 2018 - 6:12:21 AM
Long-term archiving on : Thursday, July 3, 2014 - 5:21:04 PM

File

2003-060_hal.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : ineris-00972424, version 1
  • INERIS : PU-2003-060

Collections

Citation

Billy Amzal, Eric Parent, Frédéric Y. Bois, Christian P. Robert. Optimisation de plans d'expérience par méthodes particulaires. 35. Journées de Statistiques, Jun 2003, Lyon, France. pp.97-100. ⟨ineris-00972424⟩

Share

Metrics

Record views

174

Files downloads

502