A DATA PARALLEL PROBLEM SOLVING ARCHITECTURE FOR THE RECONSTRUCTION OF PHYSICAL EVENTS
ANDREAS HEMKER · International Journal of Modern Physics C · 1993
A system is presented which is able to reconstruct the physical processes after the annihilation of particles in collider rings like LEP. It can be shown that the reconstruction task belongs to the class of NP-complete problems. Neither the conventional symbol processing approach of artificial intelligence nor connectionist systems alone satisfy the requirements of the application. A hybrid system is introduced in which a genetic algorithm acts together with a model-based hillclimbing component. The power of this method is based on the interaction of a dynamically and an inferentially working module. The inherent parallelism of dynamic systems makes a implementation on a parallel architecture useful. A data parallel implementation of this algorithm on the Connection Machine Model CM-2 is described. At last the reconstruction system is applied to a classification task, the identification of the initial quark pair. The results are compared with a conventional analysis method and a neural network.