Optimal and fast detection with the scan and with the average likelihood ratio
Guenther Walther, Stanford University
Abstract: Scan statistics are the standard tool for a range of detection problems, such as the detection of spatial disease clusters. There have been a number of recent claims in the literature, based on empirical findings,that scan statistics are inferior to an approach that involves the average likelihood ratio. The talk will look into this issue and present heuristics as well as mathematical optimality results. If time permits, a connection to efficient algorithms will be discussed.