Fast Neural Decision System Based on DSPs and Parallel Processing

A prototype of an online event validation system is developed for application in a high energy collider experiment The system mainly uses neural net works for extracting rare events with physics sig ni cance from a huge background noise It is based on processing the information collected from dif ferent detectors placed around the collision point Combining a feature extraction phase for each de tector with a global decision phase for nal deci sion on discarding or not a given event the system acts on events previously selected by a rst level analysis that reduces the event rate to kHz To cope with this input frequency the proposed system is being emulated in a node transputer based parallel machine that has a fast digital sig nal processor running as a coprocessor for each node