A randomized distributed ellipsoid algorithm for uncertain feasibility problems
Mohammadreza Chamanbaz, Giuseppe Notarstefano, Roland Bouffanais · 2017
In this paper, we consider a network of processors aiming at cooperatively solving a convex feasibility problem in which the constraint set is the intersection of local uncertain sets, each one known only by one processor. We propose a randomized, distributed method - using concepts borrowed from a centralized ellipsoid algorithm - having finite-time convergence and working under asynchronous, time-varying and directed communication topologies. At every communication round, each processor maintains a “candidate” ellipsoid for the global problem and performs two tasks. First, it verifies- in a probabilistic sense - if the center of the candidate ellipsoid is robustly feasible for its local set and, if not, constructs a new ellipsoid with smaller volume. Second, it exchanges its ellipsoid with neighbors, and then selects the one with smallest volume among the collected ones. We show that in a finite number of communication rounds, the processors reach consensus on a common ellipsoid whose center is - with high confidence- feasible for the entire set of uncertainty except a subset having an arbitrary small probability measure. We corroborate the theoretical results with numerical computations in which the algorithm is tested on a multi-core platform of processors communicating asynchronously.