From stochastic optimization to more efficient simulation algorithms

Datum

08.11.2012

Zeit

14:00 - 15:00

Sprecher

Dr. Ivo F. Sbalzarini

Zugehörigkeit

Max Planck Institute of Molecular Cell Biology and Genetics, Dresden

Serie

TUD Mathematik AG Analysis & Stochastik

Sprache

en

Hauptthema

Mathematik

Andere Themen

Mathematik

Host

Prof. Dr. R. Schilling

Beschreibung

Stochastic dynamics provides a rich framework for analyzing and designing optimization and simulation methods. In continuous black-box optimization, bio-inspired concepts from Darwinian evolution have triggered the development of numerous algorithms, including Genetic Algorithms and Evolution Strategies. Despite the success of these algorithms in real-world applications, however, the underlying stochastic dynamics remains poorly understood, and guarantees of solution quality and convergence are not available for any but the simplest toy cases. We analyze stochastic optimization algorithms as probability processes based on Gaussian Adaptation and illustrate several analogies and connections with stochastic processes in chemistry and physics, as well as with adaptive MCMC sampling. Exploiting these connections helps understand stochastic optimization algorithms, but also inspires more efficient algorithms for adaptive sampling and stochastic simulations. The latter is shown for chemical kinetics on the example of the PDM exact Stochastic Simulation Algorithm.

Links

• AG Analysis & Stochastik

Letztmalig verändert: 10.10.2012, 13:26:31