Dimension is an inherent bottleneck to some modern learning tasks, where optimization methods suffer from the size of the data. In this paper, we study non-isotropic distributions of data and develop tools that aim at reducing these dimensional costs by a dependency on an effective dimension rather than the ambient one. Based on non-asymptotic estimates of the metric entropy of ellipsoids -- that prove to generalize to infinite dimensions -- and on a chaining argument, our uniform concentration bounds involve an effective dimension instead of the global dimension, improving over existing results. We show the importance of taking advantage of non-isotropic properties in learning problems with the following applications: i) we improve state-of-the-art results in statistical preconditioning for communication-efficient distributed optimization, ii) we introduce a non-isotropic randomized smoothing for non-smooth optimization. Both applications cover a class of functions that encompasses empirical risk minization (ERM) for linear models.
Recently, local SGD has got much attention and been extensively studied in the distributed learning community to overcome the communication bottleneck problem. However, the superiority of local SGD to minibatch SGD only holds in quite limited situations. In this paper, we study a new local algorithm called Bias-Variance Reduced Local SGD (BVR-L-SGD) for nonconvex distributed optimization. Algorithmically, our proposed bias and variance reduced local gradient estimator fully utilizes small second-order heterogeneity of local objectives and suggests randomly picking up one of the local models instead of taking the average of them when workers are synchronized. Theoretically, under small heterogeneity of local objectives, we show that BVR-L-SGD achieves better communication complexity than both the previous non-local and local methods under mild conditions, and particularly BVR-L-SGD is the first method that breaks the barrier of communication complexity $\Theta(1/\varepsilon)$ for general nonconvex smooth objectives when the heterogeneity is small and the local computation budget is large. Numerical results are given to verify the theoretical findings and give empirical evidence of the superiority of our method.
In this work, we study a novel class of projection-based algorithms for linearly constrained problems (LCPs) which have a lot of applications in statistics, optimization, and machine learning. Conventional primal gradient-based methods for LCPs call a projection after each (stochastic) gradient descent, resulting in that the required number of projections equals that of gradient descents (or total iterations). Motivated by the recent progress in distributed optimization, we propose the delayed projection technique that calls a projection once for a while, lowering the projection frequency and improving the projection efficiency. Accordingly, we devise a series of stochastic methods for LCPs using the technique, including a variance reduced method and an accelerated one. We theoretically show that it is feasible to improve projection efficiency in both strongly convex and generally convex cases. Our analysis is simple and unified and can be easily extended to other methods using delayed projections. When applying our new algorithms to federated optimization, a newfangled and privacy-preserving subfield in distributed optimization, we obtain not only a variance reduced federated algorithm with convergence rates better than previous works, but also the first accelerated method able to handle data heterogeneity inherent in federated optimization.
In this work, we develop practical user scheduling algorithms for downlink bursty traffic with emphasis on user fairness. In contrast to the conventional scheduling algorithms that either equally divides the transmission time slots among users or maximizing some ratios without physcial meanings, we propose to use the 5%-tile user data rate (5TUDR) as the metric to evaluate user fairness. Since it is difficult to directly optimize 5TUDR, we first cast the problem into the stochastic game framework and subsequently propose a Multi-Agent Reinforcement Learning (MARL)-based algorithm to perform distributed optimization on the resource block group (RBG) allocation. Furthermore, each MARL agent is designed to take information measured by network counters from multiple network layers (e.g. Channel Quality Indicator, Buffer size) as the input states while the RBG allocation as action with a proposed reward function designed to maximize 5TUDR. Extensive simulation is performed to show that the proposed MARL-based scheduler can achieve fair scheduling while maintaining good average network throughput as compared to conventional schedulers.
We study adversary-resilient stochastic distributed optimization, in which $m$ machines can independently compute stochastic gradients, and cooperate to jointly optimize over their local objective functions. However, an $\alpha$-fraction of the machines are $\textit{Byzantine}$, in that they may behave in arbitrary, adversarial ways. We consider a variant of this procedure in the challenging $\textit{non-convex}$ case. Our main result is a new algorithm SafeguardSGD which can provably escape saddle points and find approximate local minima of the non-convex objective. The algorithm is based on a new concentration filtering technique, and its sample and time complexity bounds match the best known theoretical bounds in the stochastic, distributed setting when no Byzantine machines are present. Our algorithm is very practical: it improves upon the performance of all prior methods when training deep neural networks, it is relatively lightweight, and it is the first method to withstand two recently-proposed Byzantine attacks.
We present a distributed optimization algorithm for solving online personalized optimization problems over a network of computing and communicating nodes, each of which linked to a specific user. The local objective functions are assumed to have a composite structure and to consist of a known time-varying (engineering) part and an unknown (user-specific) part. Regarding the unknown part, it is assumed to have a known parametric (e.g., quadratic) structure a priori, whose parameters are to be learned along with the evolution of the algorithm. The algorithm is composed of two intertwined components: (i) a dynamic gradient tracking scheme for finding local solution estimates and (ii) a recursive least squares scheme for estimating the unknown parameters via user's noisy feedback on the local solution estimates. The algorithm is shown to exhibit a bounded regret under suitable assumptions. Finally, a numerical example corroborates the theoretical analysis.
Massive device connectivity in Internet of Thing (IoT) networks with sporadic traffic poses significant communication challenges. To overcome this challenge, the serving base station is required to detect the active devices and estimate the corresponding channel state information during each coherence block. The corresponding joint activity detection and channel estimation problem can be formulated as a group sparse estimation problem, also known under the name "Group Lasso". This letter presents a fast and efficient distributed algorithm to solve such Group Lasso problems, which alternates between solving small-scaled problems in parallel and dealing with a linear equation for consensus. Numerical results demonstrate the speedup of this algorithm compared with the state-of-the-art methods in terms of convergence speed and computation time.
This paper describes a control approach for large-scale electricity networks, with the goal of efficiently coordinating distributed generators to balance unexpected load variations with respect to nominal forecasts. To mitigate the difficulties due to the size of the problem, the proposed methodology is divided in two steps. First, the network is partitioned into clusters, composed of several dispatchable and non dispatchable generators, storage systems, and loads. A clustering algorithm is designed with the aim of obtaining clusters with the following characteristics: (i) they must be compact, keeping the distance between generators and loads as small as possible; (ii) they must be able to internally balance load variations to the maximum possible extent. Once the network clustering has been completed, a two layer control system is designed. At the lower layer, a local Model Predictive Controller is associated to each cluster for managing the available generation and storage elements to compensate local load variations. If the local sources are not sufficient to balance the cluster's load variations, a power request is sent to the supervisory layer, which optimally distributes additional resources available from the other clusters of the network. To enhance the scalability of the approach, the supervisor is implemented relying on a fully distributed optimization algorithm. The IEEE 118-bus system is used to test the proposed design procedure in a non trivial scenario.
We present the Stochastic alternate Linearization Method (StochaLM), a token-based method for distributed optimization. This algorithm finds the solution of a consensus optimization problem by solving a sequence of subproblems where some components of the cost are linearized around specific anchor points. StochaLM can be interpreted as a dual block coordinate ascent method whose block components are selected using the state of an ergodic Markov chain. This sampling process is neither essentially cyclic nor independent over time, preventing us from using proofs of convergence of dual block coordinate ascent methods done in previous works. The proof of convergence of our method is, therefore, also novel. We show that, if the cost is strongly convex and the network is fully connected, then, with probability one, the primal sequence generated by StochaLM converges to the solution. Our method is application-friendly, as it has no global hyperparameters to tune; any hyperparameters can be tuned locally by the agents, using information regarding their private cost and neighbourhood. Our method is, therefore, decentralized in the truest sense. Numerical experiments evidence that our method converges to the solution faster than other token-based methods, even when these methods' hyperparameters are tuned for optimal performance.
We propose a novel algorithm for solving convex, constrained and distributed optimization problems defined on multi-agent-networks, where each agent has exclusive access to a part of the global objective function. The agents are able to exchange information over a directed, weighted communication graph, which can be represented as a column-stochastic matrix. The algorithm combines an adjusted push-sum consensus protocol for information diffusion and a gradient descent-ascent on the local cost functions, providing convergence to the optimum of their sum. We provide results on a reformulation of the push-sum into single matrix-updates and prove convergence of the proposed algorithm to an optimal solution, given standard assumptions in distributed optimization. The algorithm is applied to a distributed economic dispatch problem, in which the constraints can be expressed in local and global subsets.
This article reports an algorithm for multi-agent distributed optimization problems with a common decision variable, local linear equality and inequality constraints and set constraints with convergence rate guarantees. \textcolor{black}{The algorithm accrues all the benefits of the Alternating Direction Method of Multipliers (ADMM) approach}. It also overcomes the limitations of existing methods on convex optimization problems with linear inequality, equality and set constraints by allowing directed communication topologies. Moreover, the algorithm can be synthesized distributively. The developed algorithm has: (i) a $O(1/k)$ rate of convergence, where $k$ is the iteration counter, when individual functions are convex but not-necessarily differentiable, and (ii) a geometric rate of convergence to any arbitrary small neighborhood of the optimal solution, when the objective functions are smooth and restricted strongly convex at the optimal solution. The efficacy of the algorithm is evaluated by a comparison with state-of-the-art constrained optimization algorithms in solving a constrained distributed $\ell_1$-regularized logistic regression problem, and unconstrained optimization algorithms in solving a $\ell_1$-regularized Huber loss minimization problem. Additionally, a comparison of the algorithm's performance with other algorithms in the literature that utilize multiple communication steps is provided.
This letter studies a distributed collision avoidance control problem for a group of rigid bodies on a sphere. A rigid body network, consisting of multiple rigid bodies constrained to a spherical surface and an interconnection topology, is first formulated. In this formulation, it is shown that motion coordination on a sphere is equivalent to attitude coordination on the 3-dimensional Special Orthogonal group. Then, an angle-based control barrier function that can handle a geodesic distance constraint on a spherical surface is presented. The proposed control barrier function is then extended to a relative motion case and applied to a collision avoidance problem for a rigid body network operating on a sphere. Each rigid body chooses its control input by solving a distributed optimization problem to achieve a nominal distributed motion coordination strategy while satisfying constraints for collision avoidance. The proposed collision-free motion coordination law is validated via simulation.
In this note, we provide an overarching analysis of primal-dual dynamics associated to linear equality-constrained optimization problems using contraction analysis. For the well-known standard version of the problem: we establish convergence under convexity and the contracting rate under strong convexity. Then, for a canonical distributed optimization problem, we use partial contractivity to establish global exponential convergence of its primal-dual dynamics. As an application, we propose a new distributed solver for the least-squares problem with the same convergence guarantees. Finally, for time-varying versions of both centralized and distributed primal-dual dynamics, we exploit their contractive nature to establish bounds on their tracking error. To support our analyses, we introduce novel results on contraction theory.
In this paper, the alternating direction method of multipliers (ADMM) is investigated for distributed optimization problems in a networked multi-agent system. In particular, a new adaptive-gain ADMM algorithm is derived in a closed form and under the standard convex property in order to greatly speed up convergence of ADMM-based distributed optimization. Using Lyapunov direct approach, the proposed solution embeds control gains into weighted network matrix among the agents and uses those weights as adaptive penalty gains in the augmented Lagrangian. It is shown that the proposed closed loop gain adaptation scheme significantly improves the convergence time of underlying ADMM optimization. Convergence analysis is provided and simulation results are included to demonstrate the effectiveness of the proposed scheme.
Many multi-agent control algorithms and dynamic agent-based models arising in natural and social sciences are based on the principle of iterative averaging. Each agent is associated to a value of interest, which may represent, for instance, the opinion of an individual in a social group, the velocity vector of a mobile robot in a flock, or the measurement of a sensor within a sensor network. This value is updated, at each iteration, to a weighted average of itself and of the values of the adjacent agents. It is well known that, under natural assumptions on the network's graph connectivity, this local averaging procedure eventually leads to global consensus, or synchronization of the values at all nodes. Applications of iterative averaging include, but are not limited to, algorithms for distributed optimization, for solution of linear and nonlinear equations, for multi-robot coordination and for opinion formation in social groups. Although these algorithms have similar structures, the mathematical techniques used for their analysis are diverse, and conditions for their convergence and differ from case to case. In this paper, we review many of these algorithms and we show that their properties can be analyzed in a unified way by using a novel tool based on recurrent averaging inequalities (RAIs). We develop a theory of RAIs and apply it to the analysis of several important multi-agent algorithms recently proposed in the literature.
In this article, we propose a new approach, optimize then agree for minimizing a sum $ f = \sum_{i=1}^n f_i(x)$ of convex objective functions over a directed graph. The optimize then agree approach decouples the optimization step and the consensus step in a distributed optimization framework. The key motivation for optimize then agree is to guarantee that the disagreement between the estimates of the agents during every iteration of the distributed optimization algorithm remains under any apriori specified tolerance; existing algorithms do not provide such a guarantee which is required in many practical scenarios. In this method, each agent during each iteration maintains an estimate of the optimal solution and, utilizes its locally available gradient information along with a finite-time approximate consensus protocol to move towards the optimal solution (hence the name Gradient-Consensus algorithm). We establish that the proposed algorithm has a global R-linear rate of convergence if the aggregate function $f$ is strongly convex and Lipschitz differentiable. We also show that under the relaxed assumption of $f_i$'s being convex and Lipschitz differentiable, the objective function error residual decreases at a Q-linear rate (in terms of the number of gradient computation steps) until it reaches a small value, which can be managed using the tolerance value specified on the finite-time approximate consensus protocol; no existing method in the literature has such strong convergence guarantees when $f_i$ are not necessarily strongly convex functions. The communication overhead for the improved guarantees on meeting constraints and better convergence of our algorithm is $O(k\log k)$ iterates in comparison to $O(k)$ of the traditional algorithms. Further, we numerically evaluate the performance of the proposed algorithm by solving a distributed logistic regression problem.
The recent deployment of distributed battery units in prosumer premises offer new opportunities for providing aggregated flexibility services to both distribution system operators and balance responsible parties. The optimization problem presented in this paper is formulated with an objective of cost minimization which includes energy and battery degradation cost to provide flexibility services. A decomposed solution approach with the alternating direction method of multipliers (ADMM) is used instead of commonly adopted centralised optimization to reduce the computational burden and time, and then reduce scalability limitations. In this work we apply a modified version of ADMM that includes two new features with respect to the original algorithm: first, the primal variables are updated concurrently, which reduces significantly the computational cost when we have a large number of involved prosumers; second, it includes a regularization term named Proximal Jacobian (PJ) that ensures the stability of the solution. A case study is presented for optimal battery operation of 100 prosumer sites with real-life data. The proposed method finds a solution which is equivalent to the centralised optimization problem and is computed between 5 and 12 times faster. Thus, aggregators or large-scale energy communities can use this scalable algorithm to provide flexibility services.
The convergence performance of distributed optimization algorithms is of significant importance to solve optimal power flow (OPF) in a distributed fashion. In this paper, we aim to provide some insights on how to partition a power system to achieve a high convergence rate of distributed algorithms for the solution of an OPF problem. We analyzed several features of the power network to find a set of suitable partitions with the aim of convergence performance improvement. We model the grid as a graph and decompose it based on the edge betweenness graph clustering. This technique provides several partitions. To find an effective partitioning, we merge the partitions obtained by clustering technique and analyze them based on characteristics of tie-lines connecting neighboring partitions. The main goal is to find the best set of partitions with respect to the convergence speed. We deploy analytical target cascading (ATC) method to distributedly solve optimization subproblems. We test the proposed algorithm on the IEEE 118-bus system. The results show that the algorithm converges faster with a proper partitioning, whereas improper partitioning leads to a large number of iterations
Distributed optimization often requires finding the minimum of a global objective function written as a sum of local functions. A group of agents work collectively to minimize the global function. We study a continuous-time decentralized mirror descent algorithm that uses purely local gradient information to converge to the global optimal solution. The algorithm enforces consensus among agents using the idea of integral feedback. Recently, Sun and Shahrampour (2020) studied the asymptotic convergence of this algorithm for when the global function is strongly convex but local functions are convex. Using control theory tools, in this work, we prove that the algorithm indeed achieves (local) exponential convergence. We also provide a numerical experiment on a real data-set as a validation of the convergence speed of our algorithm.
Communication efficient distributed mean estimation is an important primitive that arises in many distributed learning and optimization scenarios such as federated learning. Without any probabilistic assumptions on the underlying data, we study the problem of distributed mean estimation where the server has access to side information. We propose \emph{Wyner-Ziv estimators}, which are communication and computationally efficient and near-optimal when an upper bound for the distance between the side information and the data is known. As a corollary, we also show that our algorithms provide efficient schemes for the classic Wyner-Ziv problem in information theory. In a different direction, when there is no knowledge assumed about the distance between side information and the data, we present an alternative Wyner-Ziv estimator that uses correlated sampling. This latter setting offers {\em universal recovery guarantees}, and perhaps will be of interest in practice when the number of users is large and keeping track of the distances between the data and the side information may not be possible. With this mean estimator at our disposal, we revisit basic problems in decentralized optimization and compression where our Wyner-Ziv estimator yields algorithms with almost optimal performance. First, we consider the problem of communication constrained distributed optimization and provide an algorithm which attains the optimal convergence rate by exploiting the fact that the gradient estimates are close to each other. Specifically, the gradient compression scheme in our algorithm first uses half of the parties to form side information and then uses our Wyner-Ziv estimator to compress the remaining half of the gradient estimates.