Starts at: March 20, 2014 4:00 PM
Location: Scaife Hall Auditorium - Room 125
Speaker: Wilfried Gansterer
Affiliation: University of Vienna
Refreshments provided: Yes
Distributed aggregation methods, such as consensus or gossip-based approaches, have been studied intensely in the last decade and many fundamental contributions have been made. However, to the best of our knowledge, some aspects related to realistic distributed networks (no centralized control, hardly any or no global information, communication only within the local neighborhood) and silent message loss (maintaining convergence to the correct limit) have not been addressed in detail so far.
We discuss an approach for provably reliable distributed aggregation under more realistic assumptions on the distributed environment (e.g., a sensor network). By taking theoretical as well as practical points of view into account, we point out several issues concerning the practical applicability of the vast amount of existing methods and their analyses. We outline a novel convergence theory for a large class of distributed aggregation algorithms, which is more general than existing work and relies only on purely local assumptions. Based on this more general convergence theory, provably reliable distributed aggregation algorithms can be derived which successfully handle some of the challenges arising in practical computations on realistic distributed systems.
Wilfried Gansterer is an Associate Professor at the Faculty of Computer Science of the University of Vienna. He received a M.Sc. degree in Mathematics from Vienna University of Technology, a M.Sc. degree in Scientific Computing/Computational Mathematics from Stanford University, and a Ph.D. degree in Scientific Computing from Vienna University of Technology. After a post-doc stay at the Department of Computer Science at the University of Tennessee in Knoxville he joined the University of Vienna, where he received tenure in 2012. His research interests include (among others) high performance numerical computing, parallel and distributed computing, fault tolerant numerical algorithms and Internet security with a focus on botnet detection and prevention.