Robust Rapid Change-Point Detection in Multi-Sensor Data Fusion and Behavior Research

摘要

The overall goal of our AFOSR project is to develop a general and systematic foundation and methodologies for robust rapid change-point detection in the context of fusing noisy data from heterogeneous networked sensors, and apply them to model behavior data in experiments with uncertain onset time of stimulus. Specifically, we derive quickest detection schemes that are asymptotically optimal under different scenarios and spurred by the two-choice experiments in which the subject does not know the time of occurrence of the signal and is allowed to make decisions before the signal appears, we extend Ratcliff's diffusion model to the 2-CUSUM process model by emphasizing natural connections to the sequential change-point detection problems in statistics.