Anomaly detection under a nonlinear system cost objective function

摘要: We consider the problem of anomaly detection among K heterogeneous processes. At each given time, a single observation (or fixed batch observations) is collected from chosen process. The observations process follow two different distributions, depending on whether normal or abnormal. Each anomalous incurs cost until its identified and fixed, nonlinear (specifically, polynomial with degree d) duration state. objective sequential search strategy that minimizes total expected incurred by all processes during under reliability constraints. propose algorithm consists exploration, exploitation, testing phases. analyze approximation ratio regret for d > 1, establish asymptotic optimality =1.