Second-order conceptual clustering of temporal events

The characteristics of each detection of seismic activity by networks of sensors throughout the world and the time interval between detections are used to group a subset of detections, which then corresponds to a seismic event. A software system that automates the generation of descriptions that can be used to classify common seismic events is described. This learning program, called SEISMO, observes low-level signal information and forms symbolic concepts corresponding to these seismic events. These concepts are represented as scripts. The system uses probability theory for representing certainty. Thus, the use of scripts may generate multiple interpretations of a detection sequence. Ambiguities, if an exist, must be resolved by an expert seismologist. Initial experiments with the system have been positive. The system has been able to significantly offload the major portion of script generation from the domain expert. The system scanned three days worth of data in a few minutes and generated eight meaningful scripts among 13 hypothesized ones.<
>