Online distributed sensor selection

Abstract

A key problem in sensor networks is to decide which sensors to query when, in order to obtain the most useful information (e.g., for performing accurate prediction), subject to constraints (e.g., on power and bandwidth). In many applications the utility function is not known a priori, must be learned from data, and can even change over time. Furthermore for large sensor networks solving a centralized optimization problem to select sensors is not feasible, and thus we seek a fully distributed solution. In this paper, we present Distributed Online Greedy (DOG), an efficient, distributed algorithm for repeatedly selecting sensors online, only receiving feedback about the utility of the selected sensors. We prove very strong theoretical no-regret guarantees that apply whenever the (unknown) utility function satisfies a natural diminishing returns property called submodularity. Our algorithm has extremely low communication requirements, and scales well to large sensor deployments. We extend DOG to allow observationdependent sensor selection. We empirically demonstrate the effectiveness of our algorithm on several real-world sensing tasks. ABSTRACTA key problem in sensor networks is to decide which sensors to query when, in order to obtain the most useful information (e.g., for performing accurate prediction), subject to constraints (e.g., on power and bandwidth). In many applications the utility function is not known a priori, must be learned from data, and can even change over time. Furthermore for large sensor networks solving a centralized optimization problem to select sensors is not feasible, and thus we seek a fully distributed solution. In this paper, we present Distributed Online Greedy (DOG), an efficient, distributed algorithm for repeatedly selecting sensors online, only receiving feedback about the utility of the selected sensors. We prove very strong theoretical no-regret guarantees that apply whenever the (unknown) utility function satisfies a natural diminishing returns property called submodularity. Our algorithm has extremely low communication requirements, and scales well to large sensor deployments. We extend DOG to allow observationdependent sensor selection. We empirically demonstrate the effectiveness of our algorithm on several real-world sensing tasks.

Submodularity in Dynamics and Control of Networked Systems

“…To the best of our knowledge, the only existing work on distributed submodular maximization in either the online or offline case is in the context of distributed online sensor selection [56]. This approach is based on a stronger computational model, which assumes time synchronization and requires each node to compute the incremental benefit of being included in the set S t , which we do not assume.…”

“…This approach is based on a stronger computational model, which assumes time synchronization and requires each node to compute the incremental benefit of being included in the set S t , which we do not assume. Moreover, unconstrained submodular maximization is not considered in [56].…”

“…In what follows, we evaluate our proposed algorithms for constrained and unconstrained submodular maximization, and compare with the constrained submodular maximization algorithm Distributed-online-greedy (DOG) introduced in [56]. We consider the following criteria: (a) the computational assumptions of the nodes specified by the algorithms, (b) the asymptotic bounds that are achieved, (c) the rate at which the algorithms must be computed in order to reach a given value ǫ from the asymptotic bound, (d) the communication overhead, and (e) the storage overhead at each node.…”

“…Our analysis makes the same assumptions as that of [56] in order to guarantee consistency of the results, with one exception. While [56] assumed that the cost of broadcasting a message to all other nodes is 1, we assume that nodes communicate via multi-hop paths, leading to a cost of broadcast of O( n π − √ n) [21].…”

“…While [56] assumed that the cost of broadcasting a message to all other nodes is 1, we assume that nodes communicate via multi-hop paths, leading to a cost of broadcast of O( n π − √ n) [21]. Similarly, we assume that the cost of sending a message to a single node is proportional to the network diameter, i.e., O( √ n).…”