The detections and events produced by autonomous seismic data analysis systems, such as those of the NORESS and ARCESS seismic arrays, and seismic networks around the world, are a fundamental source of seismic data that underlies a wide variety of seismic research. While shortcomings in the performance of such systems may become obvious over time, remedying them can be difficult and problems with an autonomous system may persist indefinitely. This paper describes how the performance of autonomous systems can be improved over time using optimization and machine learning techniques. For example:
Optimization techniques such as Genetic Algorithms can optimize the rule thresholds of existing systems.
Supervised learning techniques such as neural networks, ID3, and CART can synthesize algorithms out of the data itself given only the levels of human supervision used in routine seismic processing. Experiments using autonomic detections from the NORESS and ARCESS array demonstrate how components of an autonomous system can be developed automatically with minimal human guidance.