Alireza Tahbaz-Salehi

Non-Bayesian social learning

We develop a dynamic model of opinion formation in social networks when the information required for learning a parameter may not be at the disposal of any single agent. Individuals engage in communication with their neighbors in order to learn from their experiences. However, instead of incorporating the views of their neighbors in a fully Bayesian manner, agents use a simple updating rule which linearly combines their personal experience and the views of their neighbors. We show that, as long as individuals take their personal signals into account in a Bayesian way, repeated interactions lead them to successfully aggregate information and learn the true parameter. This result holds in spite of the apparent naivete of agentsʼ updating rule, the agentsʼ need for information from sources the existence of which they may not be aware of, worst prior views, and the assumption that no agent can tell whether her own views or those of her neighbors are more accurate.

Consensus Over Ergodic Stationary Graph Processes

In this technical note, we provide a necessary and sufficient condition for convergence of consensus algorithms when the underlying graphs of the network are generated by an ergodic and stationary random process. We prove that consensus algorithms converge almost surely, if and only if, the expected graph of the network contains a directed spanning tree. Our results contain the case of independent and identically distributed graph processes as a special case. We also compute the mean and variance of the random consensus value that the algorithm converges to and provide a necessary and sufficient condition for the distribution of the consensus value to be degenerate.

Distributed coverage verification in sensor networks without location information

In this paper, we present three distributed algorithms for coverage verification in sensor networks with no location information. We demonstrate how, in the absence of localization devices, simplicial complexes and tools from algebraic topology can be used in providing valuable information about the properties of the cover. Our approach is based on computation of homologies of the Rips complex corresponding to the sensor network. First, we present a decentralized scheme based on Laplacian flows to compute a generator of the first homology, which represents coverage holes. Then, we formulate the problem of localizing coverage holes as an optimization problem for computing a sparse generator of the first homology. Furthermore, we show that one can detect redundancies in the sensor network by finding a sparse generator of the second homology of the cover relative to its boundary. We demonstrate how subgradient methods can be used in solving these optimization problems in a distributed manner. Finally, we provide simulations that illustrate the performance of our algorithms.

Cascades in Networks and Aggregate Volatility

We provide a general framework for the study of cascade effects created by interconnections between sectors, firms or financial institutions. Focusing on a multi sector economy linked through a supply network, we show how structural properties of the supply network determine both whether aggregate volatility disappears as the number of sectors increases (i.e., whether the law of large numbers holds) and when it does, the rate at which this happens. Our main results characterize the relationship between first order interconnections (captured by the weighted degree sequence in the graph induced by the input-output relations) and aggregate volatility, and more importantly, the relationship between higher-order interconnections and aggregate volatility. These higher-order interconnections capture the cascade effects, whereby low productivity or the failure of a set of suppliers propagates through the rest of the economy as their downstream sectors/firms also suffer and transmit the negative shock to their downstream sectors/firms. We also link the probabilities of tail events (large negative deviations of aggregate output from its mean) to sector-specific volatility and to the structural properties of the supply network.

Information Heterogeneity and the Speed of Learning in Social Networks

This paper examines how the structure of a social network and the quality of information available to different agents determine the speed of social learning. To this end, we study a variant of the seminal model of DeGroot (1974), according to which agents linearly combine their personal experiences with the views of their neighbors. We show that the rate of learning has a simple analytical characterization in terms of the relative entropy of agents’ signal structures and their eigenvector centralities. Our characterization establishes that the way information is dispersed throughout the social network has non-trivial implications for the rate of learning. In particular, we show that when the informativeness of different agents’ signal structures are comparable in the sense of Blackwell (1953), then a positive assortative matching of signal qualities and eigenvector centralities maximizes the rate of learning. On the other hand, if information structures are such that each individual possesses some information crucial for learning, then the rate of learning is higher when agents with the best signals are located at the periphery of the network. Finally, we show that the extent of asymmetry in the structure of the social network plays a key role in the long-run dynamics of the beliefs.

A One-Parameter Family of Distributed Consensus Algorithms with Boundary: From Shortest Paths to Mean Hitting Times

We present a one-parameter family of consensus algorithms over a time-varying network of agents. The proposed family of algorithms contains the average and minimum consensus algorithms as two special cases. Furthermore, we investigate a closely related family of distributed algorithms which can be considered as a consensus scheme with fixed boundary conditions and constant inputs. The proposed algorithms recover both the Bellman-Ford iteration for finding shortest paths as well as the algorithm for calculating the mean hitting time of a random walk on a graph. Finally, we demonstrate the potential utility of these algorithms for routing in adhoc networks

The Network Origins of Large Economic Downturns

This paper shows that large economic downturns may result from the propagation of microeconomic shocks over the input-output linkages across different firms or sectors within the economy. Building on the framework of Acemoglu et al. (2012), we argue that the economys input-output structure can fundamentally reshape the distribution of aggregate output, increasing the likelihood of large downturns from infinitesimal to substantial. More specifically, we show that an economy with non-trivial intersectoral input-output linkages that is subject to thin-tailed productivity shocks may exhibit deep recessions as frequently as economies that are subject to heavy-tailed shocks. Moreover, we show that in the presence of input-output linkages, aggregate volatility is not necessarily a sufficient statistic for the likelihood of large downturns. Rather, depending on the shape of the distribution of the idiosyncratic shocks, different features of the economys input-output network may be of first-order importance. Finally, our results establish that the effects of the economys input-output structure and the nature of the idiosyncratic firm-level shocks on aggregate output are not separable, in the sense that the likelihood of large economic downturns is determined by the interplay between the two.

Small world phenomenon, rapidly mixing Markov chains, and average consensus algorithms

In this paper, we demonstrate the relationship between the diameter of a graph and the mixing time of a symmetric Markov chain defined on it. We use this relationship to show that graphs with the small world property have dramatically small mixing times. Based on this result, we conclude that addition of independent random edges with arbitrarily small probabilities to a cycle significantly increases the convergence speed of average consensus algorithms, meaning that small world networks reach consensus orders of magnitude faster than a cycle. Furthermore, this dramatic increase happens for any positive probability of random edges. The same argument is used to draw a similar conclusion for the case of addition of a random matching to the cycle.

On Recurrence of Graph Connectivity in Vicsek's Model of Motion Coordination for Mobile Autonomous Agents

In this paper we complete the analysis of Vicseks model of distributed coordination among kinematic planar agents. The model is a simple discrete time heading update rule for a set of kinematic agents (or self-propelled particles as referred to by Vicsek) moving in a finite plane with periodic boundary conditions. Contrary to existing results in the literature, we do not make any assumptions on connectivity but instead prove that under the update scheme, the network of agents stays jointly connected infinitely often for almost all initial conditions, resulting in global heading alignment. Our main result is derived using a famous theorem of Hermann Weyl on equidistribution of fractional parts of sequences. We also show that the Vicsek update scheme is closely related to the Kuramoto model of coupled nonlinear oscillators.

Necessary and sufficient conditions for consensus over random independent and identically distributed switching graphs

In this paper we consider the consensus problem for stochastic switched linear dynamical systems. For discrete- time and continuous-time stochastic switched linear systems, we present necessary and sufficient conditions under which such systems reach a consensus almost surely. In the discrete-time case, our assumption is that the underlying graph of the system at any given time instance is derived from a random graph process, independent of other time instances. These graphs can be weighted, directed and with dependent edges. For the continuous-time case, we assume that the switching is governed by a Poisson point process and the graphs characterizing the topology of the system are independent and identically distributed over time. For both such frameworks, we present necessary and sufficient conditions for almost sure asymptotic consensus using simple ergodicity and probabilistic arguments. These easily verifiable conditions depend on the spectrum of the average weight matrix and the average Laplacian matrix for the discrete-time and continuous-time cases, respectively.