Saturday, 9:00 - 9:50 - Room 161
Basic principles, potential and boundaries of applicability of stochastic global optimization techniques will be discussed. It will be argued that despite huge potential of stochastic methods there are also clear boundaries on the classes of problems where these methods provide reliable answers. A significant part of the talk will be concentrated on high-dimensional global optimization problems and the so-called "curse of dimensionality". We will discuss the geometry of high-dimensional balls and cubes, very slow convergence of global random search algorithms in large-dimensional problems and poor uniformity of the so-called uniformly distributed sequences of points. Different statistical and probabilistic techniques will be considered that could be used for accelerating convergence of global random search algorithms and increasing their reliability.
Anatoly Zhigljavsky. Born in 1953. Graduated from Faculty of Mathematics, St.Petersburg State University, in 1976. PhD on applied probability in 1981. Professor of statistics at the St.Petersburg State University during 1989-1997. Since 1997: Professor, Chair in Statistics at Cardiff University. Since 2008: Director of the Centre for Optimisation and Its Applications at Cardiff University.
Author or co-author of 9 monographs on the topics of stochastic global optimization (3), time series analysis (4), optimal experimental design (1) and dynamical systems (1); editor/co-editor of 10 books or special issues of journals on the topics above, author of about 150 research papers in refereed journals, organizer of several major conferences on time series analysis, experimental design and global optimization.