Probabilistic Complex Event Triggering

Computer Science DivisionUniversity of California at Berkeley, Berkeley CA 94720, USA{daisyw,ireneos,tancau}@cs.berkeley.eduAbstract. Recently, wireless sensor deviceshave been widely deployed in various ap-plication settings (including environmentalresearch, control systems, etc.). Because ofthe inherent unreliability of sensor readings,any kind of reasoning in sensor environ-ments needs to carefully account for noise.The key goal of pcet is to build an in-frastructure that can automatically infer andreason about the probabilities of triggeredevents, using a principled probabilistic modelfor the underlying sensor data. Through suchprobabilistic reasoning, pcet can incorpo-rate uncertainly factors and make finer –grain decisions on event occurrences. Thisis achieved through the use of a BayesianNetwork to directly model and exploit cor-relations across different sensors and thedefinition of a complex – event language,which allows users / applications to createhierarchies of higher-level events. As exper-imental results verify, pcet simplifies thedevelopment process and boosts the effi-ciency of any system dealing with inher-ently uncertain data streams.