Game-Theoretic Multi-Agent Control and Network Cost Allocation Under Communication Constraints

Decentralized learning of randomization-based neural networks with centralized equivalence

This chapter introduces one of the important technologies of the application layer of network mobile information services, community structure discovery technology. As for the social attributes of mobile information services brought about by human activities, the community exploration methods for different scenarios are given from the perspective of the underlying communication network architecture. The focus of this chapter is to discover the relationship between the underlying communication architecture and the top-level community discovery to intuitively establish a community structure based on actual communication conditions. This has great benefits for discovering potential relationships and exploring unknown network structures.

Load balancing plays a critical role in a cellular network. As one kind of cellular network, Radio-over-Fibre (RoF) system can provide ubiquitous high data-rate transmissions, which has attracted many attentions, but it also suffer load unbalancing problem. In order to improve the system performance, in this paper, we propose a novel loading balance scheme in RoF system based on differential game theory. The scheme formulates the load allocated to each RAP (Radio Access Point) as a Nash Equilibrium, using non-cooperative differential game to obtain the optical load allocation of each RAP. The simulations performed show that the non-cooperative differential game algorithm is applicable and the optimal load solution can be achieved.

This paper proposes a distributed dynamics that solves the least-squares problem associated with a network system of linear algebraic equations. We consider static directed multi-agent networks. Each agent in the network has access to a private subset of the linear equations. Furthermore, we assume that agents cannot acquire any information about their "out-degrees" at any time. Under the strong connectivity condition on the underlying communication network, we show that the local estimated solution of each agent converges exponentially to the exact least-squares solution of the associated network system of linear algebraic equations.

A survey of distributed optimization

Mobile Social Network (MSN), built on inter-connected mobile devices, enables the flexibility of information exchanges of individuals within a virtual community. MSN may be self-organized and/or infrastructure-less, and the communication links may frequently fluctuate. When link qualities degrade, it remains critical to maintain the connections of important social pairs when supporting all social pairs is impossible. To achieve this goal, we propose to proactively place some reliable links (e.g., satellite links or UAV links) into the underlying communication network, so as to improve the service quality of the important social pairs, referred to as the Maintaining Social Connections (MSC) problem. We formulate the MSC problem and prove it is NP-hard. As such, we first study a special case of MSC, which is submodular and solvable with high approximation-ratio. Since the general MSC problem is not submodular, we propose an efficient approximation algorithm. Specially, by carefully choosing two submodular functions to lower/upper bound the MSC problem, we are able to prove the high approximation ratio of the proposed algorithm using these selected submodular functions. We further develop evolutionary algorithms to iteratively adjust the link placement via different exploration strategies, which also yield guaranteed approximation-ratio. Extensive evaluations based on both synthetic and real-world social network traces demonstrate the effectiveness of our proposed algorithms.

Multiagent systems are widely recognized as a method of choice for realization of distributed time-critical applications for the smart grid. However, no general solutions have been proposed for the difficult task of system development and validation, ready for deployment, which would fully account for the underlying communication network performance. We propose a novel platform designed for this purpose, which integrates a standard multiagent development framework [Java Agent Development (JADE)] and an industry standard communications network simulator (OPNET modeler). It was realized through generic extensions of the JADE framework to provide discrete event scheduling capabilities, while the OPNET modeler was extended to provide a generic method of associating the network nodes with agents running in JADE. The adopted method adheres to the high-level architecture standard. Importantly, applications developed using this platform may be deployed on the target system without manual modifications. A distributed protection application is presented and the performance is analyzed with respect to candidate agent behaviors and communication scenarios, demonstrating that the feasibility of the application critically depends on the choices made during its design and implementation.

We address a co-design of a controller and the underlying communication network in cloud-based cyber-physical systems. When communication occurs over shared resources, delays often arise that may have destabilizing effects on the closed-loop system. In order to ensure an optimal control design in the presence of such delays, not only is it useful to have a delay-aware controller but a method by which optimal assignment of delays can be imposed on the communication network links. In this paper, we propose a delay-aware stable feedback controller that judiciously accommodates the most recent state information and a machine learning (ML) based method for determining the optimal delay assignment. This ML method consists of an offline training of a neural network whose inputs are a set of selected delays and outputs are relevant performance-optimizing metrics. The resulting neural network is shown to be capable of learning the optimal delay assignment to the various links in the communication network and therefore yielding optimal performance. The proposed method is validated using a power system case study of an IEEE 68-bus, and shown to result in a notable performance improvement where in 91% of the cases a near-optimal performance can be realized.

We consider quasi-aggregative games for large populations of heterogeneous agents, whose interaction is determined by an underlying communication network. Specifically, each agent minimizes a quadratic cost function, which depends on its own strategy and on a convex combination of the strategies of its neighbors, and is subject to heterogeneous convex constraints. We suggest two distributed algorithms that can be implemented to steer the best responses of the rational agents to a Nash equilibrium configuration. The convergence of these schemes is guaranteed under different sufficient conditions depending on the matrices defining the agents' cost functions and on the communication network.

In this paper we examine the key elements determining the performance of the HPC Challenge RandomAccess benchmark on next generation supercomputers. We find that the performance of this benchmark is closely related to the bisection bandwidth of the underlying communication network, performance of integer divide operation and details of benchmark specifications such as error tolerance and permissible multi-core mapping strategies. We demonstrate that seemingly small and innocuous changes in the benchmark can lead to significantly different system performance. We also present an algorithm to optimize RandomAccess benchmark for multi-core systems. Our algorithm uses aggregation and software routing and balances the load on the cores by specializing each of the cores for one specific routing or update function. This algorithm gives approximately a factor of 3 speedup on the Blue Gene/P system which is based on quad-core nodes.

Coordinated Path Following of Multiple UAVs for Time-Critical Missions in the Presence of Time-Varying Communication Topologies

Emerging large-scale monitoring applications rely on continuous tracking of complex data-analysis queries over collections of massive, physically-distributed data streams. Thus, in addition to the space- and time-efficiency requirements of conventional stream processing (at each remote monitor site), effective solutions also need to guarantee communication efficiency (over the underlying communication network). The complexity of the monitored query adds to the difficulty of the problem -- this is especially true for nonlinear queries (e.g., joins), where no obvious solutions exist for distributing the monitor condition across sites. The recently proposed geometric method offers a generic methodology for splitting an arbitrary (non-linear) global threshold-monitoring task into a collection of local site constraints; still, the approach relies on maintaining the complete stream(s) at each site, thus raising serious efficiency concerns for massive data streams. In this paper, we propose novel algorithms for efficiently tracking a broad class of complex aggregate queries in such distributed-streams settings. Our tracking schemes rely on a novel combination of the geometric method with compact sketch summaries of local data streams, and maintain approximate answers with provable error guarantees, while optimizing space and processing costs at each remote site and communication cost across the network. One of our key technical insights for the effective use of the geometric method lies in exploiting a much lower-dimensional space for monitoring the sketch-based estimation query. Due to the complex, highly nonlinear nature of these estimates, efficiently monitoring the local geometric constraints poses challenging algorithmic issues for which we propose novel solutions. Experimental results on real-life data streams verify the effectiveness of our approach.

Due to the increased availability of low cost network technology, the use of networks to interconnect sensors, actuators and controllers is becoming widely accepted for the implementation of feedback control systems. Such type of feedback implementation, wherein the control loops are closed through a real-time network, are called network controlled systems (NCS). When implementing a NCS, the underlying communication network must provide a timely communication service, which must be adequate to fulfill the control application requirements. Therefore, the assessment of the network responsiveness to the real-time requirements of the control application is a fundamental issue. Both PROFIBUS and CAN networks are usually considered suitable to support small-scale NCS, due to their real-time capabilities. However their temporal responsiveness is highly dependent on both the timing characteristics of the supported message streams, such as its periodicity and the related message lengths. In this paper, we compare the timing properties of both CAN and PROFIBUS control networks. Basically, we assess their capability to support network controlled systems, through the evaluation of the related worst-case message's response time. A small example of a NCS is then used to assess the capability of the CAN control network to fulfill control application requirements.

We establish the O($\frac{1}{k}$) convergence rate for distributed stochastic gradient methods that operate over strongly convex costs and random networks. The considered class of methods is standard each node performs a weighted average of its own and its neighbors solution estimates (consensus), and takes a negative step with respect to a noisy version of its local functions gradient (innovation). The underlying communication network is modeled through a sequence of temporally independent identically distributed (i.i.d.) Laplacian matrices connected on average, while the local gradient noises are also i.i.d. in time, have finite second moment, and possibly unbounded support. We show that, after a careful setting of the consensus and innovations potentials (weights), the distributed stochastic gradient method achieves a (order-optimal) O($\frac{1}{k}$) convergence rate in the mean square distance from the solution. This is the first order-optimal convergence rate result on distributed strongly convex stochastic optimization when the network is random and/or the gradient noises have unbounded support. Simulation examples confirm the theoretical findings.

The effect of spatial structure on p-median results : Dominique Peeters and Isabelle Thomas Transportation Science Vol. 29, 1995, p. 366