Talk by Vahab Rostampour (University of Groningen)

Title: Distributed Data Driven Decision Making with Feasibility Performance Certificates

Abstract: In this talk, I will present a rigorous approach to distributed stochastic decision making using the so-called scenario-based approximation for large-scale networks of interconnected uncertain dynamical systems (agents) with private and common uncertainty sources. I will first show a technique to decompose a large-scale scenario program into a small-scale scenario program for each agent, and based on the existing results, provide novel guarantees to quantify the robustness of the resulting solutions for both cases of private and common uncertainties in a distributed framework. In such settings, each agent needs to communicate some information with neighbouring agents which is necessary due to the statistical learning features of the proposed setup. This inter-agent communication scheme might however arise some concerns about the agent’s private information. As a second part of my talk, I will present a novel privatised distributed framework such that each agent can share requested information while keep its local secrets. In addition, a soft communication scheme based on a set parameterization technique together with the notion of probabilistically reliable set is introduced to reduce the required communication. Such a reliability measure is incorporated into the feasibility of agent’s decisions in a probabilistic sense. The theoretical guarantees of the proposed distributed stochastic framework coincide with the centralized counterpart of both stochastic model predictive control and stochastic fault detections and isolations.