Large-Scale PDE-Constrained Optimization on Highy Parallel Multicore Architectures
Decanato - Facoltà di scienze informatiche
Data d'inizio: 15 Agosto 2010
Data di fine: 16 Agosto 2010
The Faculty of Informatics is pleased to announce a seminar given by Dr. Olaf Schenk (University of Basel)
DATE: Monday, August 16th, 2009
PLACE: USI Università della Svizzera italiana, room SI-008, Informatics Building (Via G. Buffi 13)
One of the outstanding challenges of computational sciences and engineering is large-scale nonlinear parameter estimation governed by partial differential equations. These inverse problems are known as PDE-constrained optimization problems and are significantly more difficult to solve than PDE forward problems. Ill-posedness in the optimization problem is tackled through regularization while the inclusion of inequality constraints is used to encode prior knowledge and thereby enable the algorithm to avoid many of the local minima (false solutions). The resulting nonconvex optimization problem is solved by a primal-dual interior-point algorithm with inexact step computation, which accommodates either direct factorization or iterative methods for the computation of the search direction. In the latter case we use a recently developed algebraic highly parallel method for indefinite linear systems. In addition, achieving scalability for the optimization process implemented on large-scale computing platforms comprised of tens of thousands of multicore processors is a task that offers many challenges. We will address these issues and will present numerical parallel results both in two and three space dimensions from wave equations and biomedical applications that illustrate the usefulness of the approach.
The research of Olaf Schenk concerns algorithmic and architectural problems in the field of computational mathematics, scientific computing and high-performance computing. The research has a strong emphasis on applications in computational science. From a mathematical and computer science perspective, this field requires a close interaction of numerical methods such as numerical linear algebra, nonlinear optimization and PDEs. In addition, high-performance information technology also plays an important role to get an insight into realistic applications. Olaf Schenk is an expert in the design and analysis of parallel and multi- and manycore algorithms for real-world applications on emerging architectures e.g. GPUs and Cells.
In these areas, he has published more than 50 peer-reviewed journal articles/conference contributions. As an editor, he is currently involved in the publication of the first book on Combinatorial Scientific Computing in the computational science series of Chapman\&Hall/CRC Press and on three invited book chapters on the topic of parallel computing on emerging manycore architectures and numerical linear algebra. He is an IEEE Senior Member, and a SIAM Member. He received a highly-competitive IBM Faculty Award on Cell Processors for Biomedical Hyperthermia Applications in 2008 and was one of the finalist of the International Itanium Award in 2009 in the area of computational intensive applications. He served in the program committee of several international supercomputing conferences and workshops such as, the International Conference for High Performance Computing, Networking, Storage and Analysis (SC), the International Conference on High Performance Computing (HiPC) and the Annual Swiss Workshop Series on High-Performance Computing (HPC@ETHZ/EPFL). He is also involved in ongoing collaborations and projects with industrial partners, among others with Intel and IBM Research.
HOST: Prof. Rolf Krause