Counting the number of nodes in Anonymous Dynamic Networks is enticing from an algorithmic perspective: an important computation in a restricted platform with promising applications. Starting with Michail, Chatzigiann...
详细信息
Counting the number of nodes in Anonymous Dynamic Networks is enticing from an algorithmic perspective: an important computation in a restricted platform with promising applications. Starting with Michail, Chatzigiannakis, and Spirakis [1], a flurry of papers sped up the running time guarantees from doubly-exponential to polynomial [2,3]. There is a common theme across all those works: a distinguished node is assumed to be present, because Counting cannot be solved deterministically without at least one. In the present work we study challenging questions that naturally follow: how to efficiently count with more than one distinguished node, or how to count without any distinguished node. More importantly, what is the minimal information needed about these distinguished nodes and what is the best we can aim for (count precision, stochastic guarantees, etc.) without any. We present negative and positive results to answer these questions. To the best of our knowledge, this is the first work that addresses them. (C) 2021 Elsevier Inc. All rights reserved.
This article studies the convex optimization problem with general constraints, where its global objective function is composed of the sum of local objective functions. The objective is to design a distributed algorith...
详细信息
This article studies the convex optimization problem with general constraints, where its global objective function is composed of the sum of local objective functions. The objective is to design a distributed algorithm to cooperatively resolve the optimization problem under the condition that only the information of each node's own local cost function and its neighbors' states can be obtained. To this end, the optimality condition of the researched optimization problem is developed in terms of the saddle point theory. On this basis, the corresponding continuous-time primal-dual algorithm is constructed for the considered constrained convex optimization problem under time-varying undirected and connected graphs. In the case that the parameters involved in the proposed algorithm satisfy certain inequality, the states of all nodes will reach consensus in finite time. Meanwhile, the average state is globally convergent to the optimal solution of the considered optimization problem under some mild and standard assumptions.
Graph Neural Networks (GNNs) have emerged as a very powerful and popular machine learning model for numerous application domains. Each stage of a GNN requires an aggregation (sparse matrix-matrix multiplication) and a...
详细信息
ISBN:
(纸本)9798350337662
Graph Neural Networks (GNNs) have emerged as a very powerful and popular machine learning model for numerous application domains. Each stage of a GNN requires an aggregation (sparse matrix-matrix multiplication) and a linear operation (dense matrix-matrix multiplication). Numerous efforts have addressed the development of distributed implementations for GNNs. Although efficient algorithms for distributed matrix multiplication are well known, the challenge here is the collective optimization of sequences of distributed matrix-matrix multiplications required for GNN, where many degrees of freedom also exist in the ordering of the component matrix-multiplication operations. This paper develops a new approach to distributed GNN, ReDistribution of Matrices (RDM), centered around communication-free distributed matrix-multiplication enabled by matrix redistribution between GNN stages. While the approach is applicable to the numerous algorithmic variants of GNN, the experimental evaluation focuses on GCN (Graph Convolutional Network), including both full-batch training as well as samplingbased training using GraphSAINT. Experimental evaluation with 2-layer and 3-layer GCN, using 128 or 256 hidden features, across eight sparse datasets, on a multi-GPU system with 8 GPUs shows that RDM attains a geometric mean speedup between 2x and 3.7x over two state-of-the-art multi-GPU GCN implementations, CAGNET and DGCL.
This work addresses voltage regulation problems in low-voltage distribution networks with a large penetration of solar photovoltaic inverters. We propose a fully distributed algorithm that computes solar inverters'...
详细信息
ISBN:
(纸本)9781665487788
This work addresses voltage regulation problems in low-voltage distribution networks with a large penetration of solar photovoltaic inverters. We propose a fully distributed algorithm that computes solar inverters' active and reactive power setpoints. We consider a convex surrogate of the Branch Flow Model, which is solved iteratively using the Alternating Direction Method of Multipliers (ADMM). Improved voltage regulation is achieved by using local voltage measurements. Only active power curtailment and inverter losses are minimized to maximize the inverter owner's profit. The algorithm is tested on low-voltage (LV) networks with more than 100 buses. This method outperforms a currently implemented scheme in LV networks while showing results close to perfect centralized control.
This brief investigates a distributed composite optimization problem over an undirected network with equality constraints. The optimization problem includes a smooth term and two possibly non-smooth terms. Most existi...
详细信息
This brief investigates a distributed composite optimization problem over an undirected network with equality constraints. The optimization problem includes a smooth term and two possibly non-smooth terms. Most existing results are devoted to developing distributed algorithms in a synchronous setting with a global clock, where the agents cannot proceed to the next iteration until the slowest agent completes its update. A new asynchronous distributed primal-dual forward-backward splitting algorithm (AD-PDFBS) is presented to solve this problem. Each agent can compute and communicate independently at different times, for different durations, with the information it has even if the latest information from its neighbors is not yet available. The convergence of AD-PDFBS is proved by transforming the asynchronous algorithm into a fixed-point problem utilizing the operator splitting scheme under bounded delay assumption. Experiment result confirms the effectiveness of AD-PDFBS.
作者:
Ilic, MarijaKosanic, MiroslavMIT
EECS Dept 77 Massachusetts Ave Cambridge MA 02139 USA MIT
LIDS 77 Massachusetts Ave Cambridge MA 02139 USA
In this paper we exploit the structure of dynamic models for general cyber-physical systems (CPS). Key to this structure is a unifying notion of an interaction variable between components/subsystems and the neighbouri...
详细信息
ISBN:
(纸本)9798350336825
In this paper we exploit the structure of dynamic models for general cyber-physical systems (CPS). Key to this structure is a unifying notion of an interaction variable between components/subsystems and the neighbouring sub-systems within a multi-layered CPS. The higher-level aggregate model derived in terms of unifying energy and power dynamics explicitly captures dynamic interactions of interest. In this paper this structure is utilised further to formulate a distributed interactive algorithm for simulating dynamic interactions and for aligning them according to generalised Tellegen's theorem. Proof of concept simulations are bench-marked against the centralised simulations of a simple DC circuit serving constant power load (CPL). Notably, the simulation results reflect the underlying physics and explicitly capture oscillations between neighbouring modules. This algorithm lends itself to multi-layered parallel implementation by means of minimal information exchange. As such, it supports simulating complex electric energy systems, in particular power-electronically controlled DC micro-grids, and, more generally multi-energy systems.
To accommodate low latency and computation-intensive services, such as the Internet-of-Things (IoT), 5G networks are expected to have cloud and edge computing capabilities. To this end, we consider a generic network s...
详细信息
ISBN:
(纸本)9781665464833
To accommodate low latency and computation-intensive services, such as the Internet-of-Things (IoT), 5G networks are expected to have cloud and edge computing capabilities. To this end, we consider a generic network setup where devices, performing analytics-related tasks, can partially process a task and offload its remainder to base stations, which can then reroute it to cloud and/or to edge servers. To account for the potentially unpredictable traffic demands and edge network dynamics, we formulate the resource allocation as an online convex optimization problem with service violation constraints and allow limited communication between neighboring nodes. To address the problem, we propose an online distributed (across the nodes) primal-dual algorithm and prove that it achieves sublinear regret and violation;in fact, the achieved bound is of the same order as the best known centralized alternative. Our results are further supported using the publicly available Milano dataset.
This brief focuses on the distributed resource allocation problem (RAP) for a general linear heterogeneous multi-agent system (MAS), in which each agent adopts different structure parameters. By means of multi-agent c...
详细信息
This brief focuses on the distributed resource allocation problem (RAP) for a general linear heterogeneous multi-agent system (MAS), in which each agent adopts different structure parameters. By means of multi-agent consensus approach and symbolic-function-based fixed-time control theory, an initialization-free distributed resource allocation algorithm is designed. Moreover, the equality constraint is solved based on the proportional-integral (PI) control idea and the output of all agents tracks to the optimal solution within fixed-time. Finally, we reveal the effectiveness and fast convergence performance of our proposed algorithm by a simulation example.
This article deals with a distributed Mixed-Integer Linear Programming (MILP) setup arising in several control applications. Agents of a network aim to minimize the sum of local linear cost functions subject to both i...
详细信息
This article deals with a distributed Mixed-Integer Linear Programming (MILP) setup arising in several control applications. Agents of a network aim to minimize the sum of local linear cost functions subject to both individual constraints and a linear coupling constraint involving all the decision variables. A key, challenging feature of the considered setup is that some components of the decision variables must assume integer values. The addressed MILPs are NP-hard, nonconvex, and large-scale. Moreover, several additional challenges arise in a distributed framework due to the coupling constraint, so that feasible solutions with guaranteed suboptimality bounds are of interest. We propose a fully distributed algorithm based on a primal decomposition approach and an appropriate tightening of the coupling constraint. The algorithm is guaranteed to provide feasible solutions in finite time. Moreover, asymptotic and finite-time suboptimality bounds are established for the computed solution. Monte Carlo simulations highlight the extremely low suboptimality bounds achieved by the algorithm.
In this work, we introduce lozenge P-vCube, a push-based failure detector for asynchronous distributed systems. Processes running lozenge P-vCube are hierarchically organized on the vCube virtual topology, which prese...
详细信息
ISBN:
(纸本)9798400708442
In this work, we introduce lozenge P-vCube, a push-based failure detector for asynchronous distributed systems. Processes running lozenge P-vCube are hierarchically organized on the vCube virtual topology, which presents several logarithmic properties. As there are no bounds on communication delays and process execution times, false suspicions may occur. To ensure eventual accuracy, a process exits the system when it learns that it has been suspected by another process or suspects every other process. We show that lozenge P-vCube guarantees completeness in any situation, and eventual accuracy if all correct processes remain strongly connected among themselves. We implemented the proposed algorithm using simulation and presented comparison results with the traditional all-test-all solution and a ring approach. The results demonstrate the trade-off between the speed of detecting actual failures and the number of messages used, as well as the scalability of lozenge P-vCube.
暂无评论