25.07.2012, 15:00 Uhr
Forschungsgruppe Theory and Applications of Algorithms
Lenaugasse 2, 1. Stock
One of the most frequent operations in the field of image processing is the matrix multiplication. Parallel algorithms and concepts are well known and developed, but in case of decentralized systems there is still a lack of new approaches. In this thesis we proposed different approaches of performing a matrix multiplication on a decentralized mobile sensor network. The aggregation process over the nodes is based on the Push Sum algorithm. We introduced four strategies for the matrix multiplication on a mobile sensor
network. Results have shown that it is relatively hard for a random node distribution and constant mobility to achieve high accuracy. Therefore, we considered different deterministic topologies with high immobility periods, particularly in the transient phase of the aggregation process. Under certain circumstances, we reduced the message loss and achieved nearly half of double precision.