Online distributed stochastic learning algorithm for convex optimization in time-varying directed networks
Introduction
In recent years, distributed optimization has been receiving a lot of attention in information sciences and engineering fields, such as machine learning, control applications for wireless networks, signal processing, and power networks and sensor networks, e.g., see [1], [2], [3], [4], [5], [6], [7], [8], [9]. Compared to centralized optimization, the distributed optimization has an essential difference characterized by the lack of full knowledge about the overall problem structure. This means that all agents of a network collectively minimize an optimization problem by local information exchange and without any centralized coordination, and each agent is only allowed to exchange the information from its immediate neighbors over the network. The underlying communication among agents can be modeled as a undirected or directed time-varying graph. The main task of distributed optimization is to collaboratively minimize the sum of several local cost functions in general, where individual agent holds a private copy of one specific cost function, see [10], for more details. As well-known, there are many recent researches focusing on distributed optimization problems and their applications, see [11], [12], [13], [14], [15].
However, many practical scenarios in distributed optimization frequently encounter dynamically changing and uncertain environments. For examples, observations are time-changing due to noises in parameter estimation problems by using sensor networks, and uncertainties play an important role in the schedule of renewable energy in power systems. To address some of these issues, the online optimization is effective to deal with uncertainties arising in these problems. Different from distributed optimization, online optimization is to minimize a time-vary cost function, and design a online algorithm to reduce the so-called regret, which measures the gap between the accumulated collective cost and the cost obtained by the best single hindsight decision made by a hypothetical decision maker knowing the cost functions in advance. A online algorithm can be claimed “good” when the regret is sub-linear [16]. Online optimization problems have been studied extensively in literatures. For solving a online convex optimization, Zinkevich et al. [16] firstly proposed a gradient-based method and gave a regret bound with order of where T is the time horizon. Later, Hazan et al. [17] obtained an improved regret rate of for twice differentiable strongly convex functions. For more results on online optimization, please refer to the survey [18] and references therein.
However, the design for most of online algorithms stated above is based on centralized architectures. Until recently, motivated the interest in decentralized optimization and its many applications, distributed versions of online optimization are developed, see [19], [20]. In a online distributed optimization, the global cost function associated a multi-agent network is represented as the sum of local cost functions, where each local function is allocated to a agent varying possibly over time, and this change is seen by agents only in hindsight. The goal is to design a online distributed algorithm that cooperatively minimize the global cost function across a time horizon. By contrasts, online distributed optimization inherently differs from distributed or online optimization [21]. Based on consensus schemes, gradient descent methods for online distributed optimization were proposed in [22], [23], where all agents are able to collaboratively reduce their average regret. The authors in [22] proposed a consensus based dual averaging algorithm over undirected graphs. Motivated by the saddle-point dynamics in [24], subgradient-based online distributed algorithm was proposed on weight-balanced network topologies [23]. Based on the assumption that communication weight matrices are double stochastic, Tsianos et al. [25] introduced a gossip-based protocol for online distributed convex optimization. Later, Hosseini et al. [26] extended their previous results [22] to accommodate for time-varying weights, but on a fixed directed graph.
Nevertheless, most of the works cited above assumed that communications among agents are either fixed or undirected. Moreover, due to uncertainties in reality, there exist noises in the evaluation of (sub)gradients when designed algorithms. Recently, Akbari et al. [21] generalized the gradient push-sum algorithm in [27] to online setting, and proposed a discrete time online distributed algorithm over time-varying directed graphs. In [28], Lee et al. investigated a online distributed optimization problem with coupled inequality constraints and designed a primal-dual online distributed algorithm. When the calculation of gradients existed noises, a decentralized stochastic variant of dual averaging methods was proposed in [29].
Motivated by recent works [27], [29], we propose a online distributed stochastic (sub)gradient push-sum algorithm in this paper. The idea of our method is based push-sum protocol on imbalanced directed graphs [27], [30]. Each agent is required to know its out-degree at each time, without requiring knowledge of either the number of agents or the graph sequence. Meanwhile, the noisy (sub)gradients are also taken into consideration. For the proposed algorithm, we obtain the regret bounds on the cases of convex and strongly convex cost functions, respectively. By contrasts, we extend the methods in [27], [30] to online setting, and explicitly give regret estimates. Compared with the work in [21], we improve the bound of regret from to for strongly convex objective functions while taking noisy gradients into consideration.
The remainder of this paper is organized as follows. In Section 2, we state the related problem, useful assumptions and algorithm. In Section 3, we give regret analysis and obtain main results. Numerical simulations are given in Section 4. Finally, Section 5 draws some conclusions.
Section snippets
Problem formulation and preliminaries
Now we state the online distributed optimization problem under consideration in this paper. Consider a time-varying multi-agent network with m agents, in which uncertainties are modeled as a sequence of time-varying objective functions unknown in advance. To be specific, at each time an agent chooses its action . After this, a cost function is revealed, and the agent incurs the cost . Hence, at each time t, the networked cost function is given by
Regret analysis: the case when cost functions are convex
In this section, we first consider the case when the cost functions are convex. In this setting, we obtain the results on regret bound for the proposed Algorithm 1. In what follows, we give a lemma to bound upper the networked pseudo-regret. Lemma 1 Suppose that Assumptions 1, 2 and 3 hold. Let be the sequence generated by Algorithm 1 with the learning rate β[t]. Define the average . Then, we have for all T ≥ 1 andx* ∈ X*
Simulation results
In this section, a numerical example on the localization of sensor networks is used to illustrate the performance of Algorithm 1. In the simulations, the average individual pseudo-regret over time Rj(T)/T is monitored as the metric of convergence.
Considering a network of m sensors, the goal is to estimate a vector . At each time each sensor i receives an observation vector . Due to observation noises, the observation vector hi[t] is time-varying. Assuming that each sensor i
Conclusion
We have investigated a online optimization problem for multi-agent systems over time-varying directed networks. We proposed a online distributed stochastic (sub)gradient push-sum algorithm by utilizing distributed optimization methods and broadcast-based push-sum protocols. Then we analyzed the pseudo-regret bounds on the convex and strongly convex cases for the proposed algorithm. Numerical experiments on the localization in sensor networks demonstrated that the effectiveness of the proposed
Declaration of interests
None.
Acknowledgments
This research was partially supported by the NSFC under grants 11501070, 11671062 and 11871128, and by the Natural Science Foundation Projection of Chongqing under grants cstc2017jcyjAX0253 and cstc2018jcyjAX0172, and the Science and Technology Research Program of Chongqing Municipal Education Commission (Grant no. KJQN201800520).
Jueyou Li received the B.E. and M.E. degrees in Mathematical and Software Science from Sichuan Normal University, Chengdu, China, in 2003 and 2006, respectively, and the Ph.D. degree in Operation Research from Federation University Australia, Australia, in 2014. He was a postdoctoral research fellow at the School of Electrical and Information Engineering, The University of Sydney, Australia, in 2015. He is now a Professor at the School of Mathematical Sciences, Chongqing Normal University,
References (30)
- et al.
Stability of periodic solution in fuzzy BAM neural networks with finite distributed delays
Neurocomputing
(2008) - et al.
Distributed mirror descent method for multi-agent optimization with delay
Neurocomputing
(2016) - et al.
Impulsive effects on stability of high-order BAM neural networks with time delays
Neurocomputing
(2011) - et al.
Distributed Control of Robotic Networks: A Mathematical Approach to Motion Coordination Algorithms
(2009) - et al.
An adaptive projected subgradient approach to learning in diffusion networks
IEEE Trans. Signal Process.
(2009) - et al.
Distributed consensus algorithms in sensor networks: Quantized data and random link failures
IEEE Trans. Signal Process.
(2010) - et al.
Distributed constrained optimization by consensus-based primal-dual perturbation method
IEEE Trans. Autom. Control
(2014) - et al.
Distributed event-triggered scheme for economic dispatch in smart grids
IEEE Trans. Ind. Inform.
(2016) - et al.
Distributed optimal consensus over resource allocation network and its application to dynamical economic dispatch
IEEE Trans. Neural Netw. Learn. Syst.
(2018)
Noncooperative game-based distributed charging control for plug-in electric vehicles in distribution networks
IEEE Trans. Ind. Inform.
Distributed subgradient methods for multi-agent optimization
IEEE Trans. Autom. Control
Constrained consensus and optimization in multi-agent networks
IEEE Trans. Autom. Control
Gradient-free method for nonsmooth distributed optimization
J. Global Optim.
Efficient computation for sparse load shifting in demand side management
IEEE Trans. Smart Grid
Cited by (8)
Distributed online adaptive subgradient optimization with dynamic bound of learning rate over time-varying networks
2022, IET Control Theory and ApplicationsDistributed adaptive online learning for convex optimization with weight decay
2022, Asian Journal of ControlAdaptive quantized online distributed stochastic mirror descent algorithm
2022, Proceedings - 2022 37th Youth Academic Annual Conference of Chinese Association of Automation, YAC 2022
Jueyou Li received the B.E. and M.E. degrees in Mathematical and Software Science from Sichuan Normal University, Chengdu, China, in 2003 and 2006, respectively, and the Ph.D. degree in Operation Research from Federation University Australia, Australia, in 2014. He was a postdoctoral research fellow at the School of Electrical and Information Engineering, The University of Sydney, Australia, in 2015. He is now a Professor at the School of Mathematical Sciences, Chongqing Normal University, China. His current research interests include optimization and control, distributed optimization, online optimization and their applications.
Chuanye Gu received the B.E. and M.E. degrees in Mathematics and Applied Mathematics from Chongqing Normal University, Chongqing, China, in 2015 and 2018, respectively. She is now a Research Assistant at the Faculty of Science & Engineering, Curtin University, Australia. Her research interests include distributed optimization, online optimization, complex network and their applications in smart grid.
Zhiyou Wu received the Ph.D. degree in Operations Research from Shanghai University, China, in 2003. She is currently a Professor and Head of School of Mathematical Sciences, Chongqing Normal University, China. Previously, she was Associate Professor at Federation University Australia, Ballarat, Australia. Her current research interests include optimization, nonlinear programming and their applications in engineering.