An Efficient Randomized Approximation Algorithm for Volume Estimation and Design Centering
Verteidigung im Promotionsverfahren von Dipl.-Biomath. Josefine Asmus
28.4.2017, 13:00 Uhr, APB 1004 (Ratssaal)
The design of systems or models that work robustly under uncertainty and environmental fluctuations is a key challenge in both engineering and science. This is formalized in the design centering problem, defined as finding a design that fulfills given specifications and has a high probability of still doing so if the system parameters or the specifications randomly fluctuate. Design centering is often accompanied by the problem of quantifying the robustness of a system. Here we present a novel adaptive statistical method to simultaneously address both problems. Our method, Lp-Adaptation, is inspired by how robustness evolves in biological systems and by randomized schemes for convex volume computation. It is able to address both problems in the general, non-convex case and at low computational cost. In this thesis, we describe the concepts of the algorithm and detail its steps. We then test it on known benchmarks, and demonstrate its real-wold applicability in electronic and biological systems. In all cases, the present method outperforms the previous state of the art. This enables re-formulating optimization problems in engineering and biology as design centering problems, taking global system robustness into account.
A C++ based MPI-enabled Tasking Framework to Efficiently Parallelize Fast Multipode Methods for
Vortrag im Promotionsverfahren (alle Studiengänge) von David Haensel (ZIH)
2.5.2017, 11:00 Uhr, APB 1004 (Ratssaal)
Modern molecular dynamic simulations steadily shift the scaling bottleneck from computation towards
communication. The main bottleneck of such simulations are long-range interactions that cannot be
computed from local information only. Any algorithm computing the equivalent of such long-range
pairwise interactions needs global information by design which leads to global communication.
Especially for strong scaling with only a few particles per core such Coulomb solvers are already
communication;ynchronization-bound. To tackle these issues we are developing a Fast Multipole
Method (FMM) in modern C++.
In this talk we will cover two main areas of activity, namely intra-node and inter-node parallelization.
The C++ language standard (C++11) offers several robust features for parallel intra-node
programming. With the help of those standardized C++ features, we added a flexible algorithm-aware
tasking framework to our FMM library. To reduce intra-node parallelization overhead, e.g.
synchronization points or load imbalance, the tasking framework provides algorithm-aware features
like multiqueues, work-stealing and compile-time configuration of dependency-resolving via template
The inter-node parallelization mainly focuses on the reduction of latency. Therefore, we have to
change the communication pattern especially along the critical communication path. Latency
avoidance can be done by partial or full replication of data, which might lead to a controlled increase of
computation for specific parts of the algorithm. To reduce the visibility of such communication
primitives (e.g. MPI) and pre- and postprocessing steps (data serialization, etc.) inside the algorithm
we introduced several abstraction layers. This allows an easy exchange of underlying communication
libraries without changing code at the algorithmic level.
In this talk we will present current results for both intra- and inter-node parallelization concepts and
provide an outlook towards unifying both concepts in an MPI-enabled tasking framework for the FMM.
Suche im Ankündigungsarchiv
Abonnieren Sie die Vortragsankündigungen als News Feed: