Optimal Distributional Design for External Stochastic Knowledge Processing

Optimal processing of distributed knowledge is key agenda in machine learning, signal processing and control, driven by sensor networks for smart environments, autonomous agents and distributed infrastruktures (clouds, Internet) serving the tnternet of things. Nodes may communicate via partially specied probability distributions (moments, etc.). If a remote node or central coordinator is to process this a global stochastic dependencymodel is required, but rarely available. Long- established entropy methods are used for distributed decision making extended, as Fully Probabilistic Design (FPD) in recent years. Restrictive modelling assumptions - which assume nodes are finitely parametrized - need to be relaxed if progress to mature applications is to be archieved. A consistent procedure for nodes to assing weight to remotely sourced knowledge is also a priority. Applications in recursive signal processing and distributed (Kalman) filtering will consequently be delivered, allowing randomized exploration of the design space, and equipping decisions with uncentrainty quantifiers.