Δ-Dataflow Networks for Event Stream Processing

We present the design of a new class of dataflow-like networks suitable for detecting complex conditions in systems where parameters change rapidly. Such networks are helpful for detecting conditions that signal threats or op-
portunities in areas such as logistics, finance, and public health. Examples of such applications are detection of money laundering, epidemics, and unauthorized intrusion into systems. We call these networks ∆-dataflow networks because nodes in the network propagate only changes in data values. We show how ultra low power asynchronous architectures that have been developed for sensor networks can provide an extremely efficient platform for executing such networks.