Current browse context:
stat.CO
Change to browse by:
References & Citations
Statistics > Computation
Title: Particle algorithms for optimization on binary spaces
(Submitted on 2 Nov 2011 (v1), last revised 6 Apr 2012 (this version, v2))
Abstract: We discuss a unified approach to stochastic optimization of pseudo-Boolean objective functions based on particle methods, including the cross-entropy method and simulated annealing as special cases. We point out the need for auxiliary sampling distributions, that is parametric families on binary spaces, which are able to reproduce complex dependency structures, and illustrate their usefulness in our numerical experiments. We provide numerical evidence that particle-driven optimization algorithms based on parametric families yield superior results on strongly multi-modal optimization problems while local search heuristics outperform them on easier problems.
Submission history
From: Christian Schäfer [view email][v1] Wed, 2 Nov 2011 17:32:23 GMT (62kb,D)
[v2] Fri, 6 Apr 2012 16:42:56 GMT (372kb,D)
Link back to: arXiv, form interface, contact.