arXiv subjects

Elchanan Mossel

Publications and source records attributed to Elchanan Mossel.

At least 163 records · Page 9Linked to original sources

arXiv2013

Robust Optimality of Gaussian Noise Stability

We prove that under the Gaussian measure, half-spaces are uniquely the most noise stable sets. We also prove a quantitative version of uniqueness, showing that a set which is almost optimally noise stable must be close to a half-space. This extends a theorem of Borell, who proved the same result but without uniqueness, and it also answers a question of Ledoux, who asked whether it was possible to prove Borell's theorem using a direct semigroup argument. Our quantitative uniqueness result has various applications in diverse fields.

math.PR↗

arXiv2013

Exact thresholds for Ising-Gibbs samplers on general graphs

We establish tight results for rapid mixing of Gibbs samplers for the Ferromagnetic Ising model on general graphs. We show that if \[(d-1)\tanhβ<1,\] then there exists a constant C such that the discrete time mixing time of Gibbs samplers for the ferromagnetic Ising model on any graph of n vertices and maximal degree d, where all interactions are bounded by $β$, and arbitrary external fields are bounded by $Cn\log n$. Moreover, the spectral gap is uniformly bounded away from 0 for all such graphs, as well as for infinite graphs of maximal degree d. We further show that when $d\tanhβ<1$, with high probability over the Erdos-Renyi random graph $G(n,d/n)$, it holds that the mixing time of Gibbs samplers is \[n^{1+Θ({1}/{\log\log n})}.\] Both results are tight, as it is known that the mixing time for random regular and Erdos-Renyi random graphs is, with high probability, exponential in n when $(d-1)\tanhβ>1$, and $d\tanhβ>1$, respectively. To our knowledge our results give the first tight sufficient conditions for rapid mixing of spin systems on general graphs. Moreover, our results are the first rigorous results establishing exact thresholds for dynamics on random graphs in terms of spatial thresholds on trees.

math.PR↗

arXiv2012

On reverse hypercontractivity

We study the notion of reverse hypercontractivity. We show that reverse hypercontractive inequalities are implied by standard hypercontractive inequalities as well as by the modified log-Sobolev inequality. Our proof is based on a new comparison lemma for Dirichlet forms and an extension of the Strook-Varapolos inequality. A consequence of our analysis is that {\em all} simple operators $L=Id-\E$ as well as their tensors satisfy uniform reverse hypercontractive inequalities. That is, for all $q<p<1$ and every positive valued function $f$ for $t \geq \log \frac{1-q}{1-p}$ we have $\| e^{-tL}f\|_{q} \geq \| f\|_{p}$. This should be contrasted with the case of hypercontractive inequalities for simple operators where $t$ is known to depend not only on $p$ and $q$ but also on the underlying space. The new reverse hypercontractive inequalities established here imply new mixing and isoperimetric results for short random walks in product spaces, for certain card-shufflings, for Glauber dynamics in high-temperatures spin systems as well as for queueing processes. The inequalities further imply a quantitative Arrow impossibility theorem for general product distributions and inverse polynomial bounds in the number of players for the non-interactive correlation distillation problem with $m$-sided dice.

math.PR↗

arXiv2012

Phylogenetic mixtures: Concentration of measure in the large-tree limit

The reconstruction of phylogenies from DNA or protein sequences is a major task of computational evolutionary biology. Common phenomena, notably variations in mutation rates across genomes and incongruences between gene lineage histories, often make it necessary to model molecular data as originating from a mixture of phylogenies. Such mixed models play an increasingly important role in practice. Using concentration of measure techniques, we show that mixtures of large trees are typically identifiable. We also derive sequence-length requirements for high-probability reconstruction.

math.PR↗

arXiv2012

From Agreement to Asymptotic Learning

We consider a group of Bayesian agents who are each given an independent signal about an unknown state of the world, and proceed to communicate with each other. We study the question of asymptotic learning: do agents learn the state of the world with probability that approaches one as the number of agents tends to infinity? We show that under general conditions asymptotic learning follows from agreement on posterior actions or posterior beliefs, regardless of the communication dynamics. In particular, we prove that asymptotic learning holds for the Gale-Kariv model on undirected networks and non-atomic private beliefs.

math.ST↗

arXiv2012

Connectivity and equilibrium in random games

We study how the structure of the interaction graph of a game affects the existence of pure Nash equilibria. In particular, for a fixed interaction graph, we are interested in whether there are pure Nash equilibria arising when random utility tables are assigned to the players. We provide conditions for the structure of the graph under which equilibria are likely to exist and complementary conditions which make the existence of equilibria highly unlikely. Our results have immediate implications for many deterministic graphs and generalize known results for random games on the complete graph. In particular, our results imply that the probability that bounded degree graphs have pure Nash equilibria is exponentially small in the size of the graph and yield a simple algorithm that finds small nonexistence certificates for a large family of graphs. Then we show that in any strongly connected graph of n vertices with expansion $(1+Ω(1))\log_2(n)$ the distribution of the number of equilibria approaches the Poisson distribution with parameter 1, asymptotically as $n \to +\infty$.

math.PR↗

arXiv2012

Majority is Stablest : Discrete and SoS

The Majority is Stablest Theorem has numerous applications in hardness of approximation and social choice theory. We give a new proof of the Majority is Stablest Theorem by induction on the dimension of the discrete cube. Unlike the previous proof, it uses neither the "invariance principle" nor Borell's result in Gaussian space. The new proof is general enough to include all previous variants of majority is stablest such as "it ain't over until it's over" and "Majority is most predictable". Moreover, the new proof allows us to derive a proof of Majority is Stablest in a constant level of the Sum of Squares hierarchy.This implies in particular that Khot-Vishnoi instance of Max-Cut does not provide a gap instance for the Lasserre hierarchy.

cs.CC↗

arXiv2012

On extracting common random bits from correlated sources on large alphabets

Suppose Alice and Bob receive strings $X=(X_1,...,X_n)$ and $Y=(Y_1,...,Y_n)$ each uniformly random in $[s]^n$ but so that $X$ and $Y$ are correlated . For each symbol $i$, we have that $Y_i = X_i$ with probability $1-\eps$ and otherwise $Y_i$ is chosen independently and uniformly from $[s]$. Alice and Bob wish to use their respective strings to extract a uniformly chosen common sequence from $[s]^k$ but without communicating. How well can they do? The trivial strategy of outputting the first $k$ symbols yields an agreement probability of $(1 - \eps + \eps/s)^k$. In a recent work by Bogdanov and Mossel it was shown that in the binary case where $s=2$ and $k = k(\eps)$ is large enough then it is possible to extract $k$ bits with a better agreement probability rate. In particular, it is possible to achieve agreement probability $(k\eps)^{-1/2} \cdot 2^{-k\eps/(2(1 - \eps/2))}$ using a random construction based on Hamming balls, and this is optimal up to lower order terms. In the current paper we consider the same problem over larger alphabet sizes $s$ and we show that the agreement probability rate changes dramatically as the alphabet grows. In particular we show no strategy can achieve agreement probability better than $(1-\eps)^k (1+δ(s))^k$ where $δ(s) \to 0$ as $s \to \infty$. We also show that Hamming ball based constructions have {\em much lower} agreement probability rate than the trivial algorithm as $s \to \infty$. Our proofs and results are intimately related to subtle properties of hypercontractive inequalities.

cs.IT↗

arXiv2012

Exit time tails from pairwise decorrelation in hidden Markov chains, with applications to dynamical percolation

Consider a Markov process ω_t at equilibrium and some event C (a subset of the state-space of the process). A natural measure of correlations in the process is the pairwise correlation \Pr[ω_0,ω_t \in C] - \Pr[ω_0 \in C]^2. A second natural measure is the probability of the continual occurrence event \{ω_s \in C, \forall s\in [0,t]\}. We show that for reversible Markov chains, and any event C, pairwise decorrelation of the event C implies a decay of the probability of the continual occurrence event \{ω_s \in C, \forall s \in [0,t]\} as t\to\infty. We provide examples showing that our results are often sharp. Our main applications are to dynamical critical percolation. Let C be the left-right crossing event of a large box, and let us scale time so that the expected number of changes to C is order 1 in unit time. We show that the continual connection event has superpolynomial decay. Furthermore, on the infinite lattice without any time scaling, the first exceptional time with an infinite cluster appears with an exponential tail.

math.PR↗

arXiv2012

Stochastic Block Models and Reconstruction

The planted partition model (also known as the stochastic blockmodel) is a classical cluster-exhibiting random graph model that has been extensively studied in statistics, physics, and computer science. In its simplest form, the planted partition model is a model for random graphs on $n$ nodes with two equal-sized clusters, with an between-class edge probability of $q$ and a within-class edge probability of $p$. Although most of the literature on this model has focused on the case of increasing degrees (ie.\ $pn, qn \to \infty$ as $n \to \infty$), the sparse case $p, q = O(1/n)$ is interesting both from a mathematical and an applied point of view. A striking conjecture of Decelle, Krzkala, Moore and Zdeborová based on deep, non-rigorous ideas from statistical physics gave a precise prediction for the algorithmic threshold of clustering in the sparse planted partition model. In particular, if $p = a/n$ and $q = b/n$, then Decelle et al.\ conjectured that it is possible to cluster in a way correlated with the true partition if $(a - b)^2 > 2(a + b)$, and impossible if $(a - b)^2 < 2(a + b)$. By comparison, the best-known rigorous result is that of Coja-Oghlan, who showed that clustering is possible if $(a - b)^2 > C (a + b)$ for some sufficiently large $C$. We prove half of their prediction, showing that it is indeed impossible to cluster if $(a - b)^2 < 2(a + b)$. Furthermore we show that it is impossible even to estimate the model parameters from the graph when $(a - b)^2 < 2(a + b)$; on the other hand, we provide a simple and efficient algorithm for estimating $a$ and $b$ when $(a - b)^2 > 2(a + b)$. Following Decelle et al, our work establishes a rigorous connection between the clustering problem, spin-glass models on the Bethe lattice and the so called reconstruction problem. This connection points to fascinating applications and open problems.

math.PR↗

arXiv2012

Asymptotic Learning on Bayesian Social Networks

Understanding information exchange and aggregation on networks is a central problem in theoretical economics, probability and statistics. We study a standard model of economic agents on the nodes of a social network graph who learn a binary "state of the world" S, from initial signals, by repeatedly observing each other's best guesses. Asymptotic learning is said to occur on a family of graphs G_n = (V_n, E_n), with |V_n| tending to infinity, if with probability tending to 1 as n tends to infinity all agents in G_n eventually estimate S correctly. We identify sufficient conditions for asymptotic learning and contruct examples where learning does not occur when the conditions do not hold.

math.PR↗

arXiv2012

Majority Dynamics and Aggregation of Information in Social Networks

Consider n individuals who, by popular vote, choose among q >= 2 alternatives, one of which is "better" than the others. Assume that each individual votes independently at random, and that the probability of voting for the better alternative is larger than the probability of voting for any other. It follows from the law of large numbers that a plurality vote among the n individuals would result in the correct outcome, with probability approaching one exponentially quickly as n tends to infinity. Our interest in this paper is in a variant of the process above where, after forming their initial opinions, the voters update their decisions based on some interaction with their neighbors in a social network. Our main example is "majority dynamics", in which each voter adopts the most popular opinion among its friends. The interaction repeats for some number of rounds and is then followed by a population-wide plurality vote. The question we tackle is that of "efficient aggregation of information": in which cases is the better alternative chosen with probability approaching one as n tends to infinity? Conversely, for which sequences of growing graphs does aggregation fail, so that the wrong alternative gets chosen with probability bounded away from zero? We construct a family of examples in which interaction prevents efficient aggregation of information, and give a condition on the social network which ensures that aggregation occurs. For the case of majority dynamics we also investigate the question of unanimity in the limit. In particular, if the voters' social network is an expander graph, we show that if the initial population is sufficiently biased towards a particular alternative then that alternative will eventually become the unanimous preference of the entire population.

math.ST↗

arXiv2012

Geometric Influences II: Correlation Inequalities and Noise Sensitivity

In a recent paper, we presented a new definition of influences in product spaces of continuous distributions, and showed that analogues of the most fundamental results on discrete influences, such as the KKL theorem, hold for the new definition in Gaussian space. In this paper we prove Gaussian analogues of two of the central applications of influences: Talagrand's lower bound on the correlation of increasing subsets of the discrete cube, and the Benjamini-Kalai-Schramm (BKS) noise sensitivity theorem. We then use the Gaussian results to obtain analogues of Talagrand's bound for all discrete probability spaces and to reestablish analogues of the BKS theorem for biased two-point product spaces.

math.PR↗

arXiv2012

Geometric influences

We present a new definition of influences in product spaces of continuous distributions. Our definition is geometric, and for monotone sets it is identical with the measure of the boundary with respect to uniform enlargement. We prove analogs of the Kahn-Kalai-Linial (KKL) and Talagrand's influence sum bounds for the new definition. We further prove an analog of a result of Friedgut showing that sets with small "influence sum" are essentially determined by a small number of coordinates. In particular, we establish the following tight analog of the KKL bound: for any set in $\mathbb{R}^n$ of Gaussian measure $t$, there exists a coordinate $i$ such that the $i$th geometric influence of the set is at least $ct(1-t)\sqrt{\log n}/n$, where $c$ is a universal constant. This result is then used to obtain an isoperimetric inequality for the Gaussian measure on $\mathbb{R}^n$ and the class of sets invariant under transitive permutation group of the coordinates.

math.PR↗

arXiv2012

Bundling Customers: How to Exploit Trust Among Customers to Maximize Seller Profit

We consider an auction of identical digital goods to customers whose valuations are drawn independently from known distributions. Myerson's classic result identifies the truthful mechanism that maximizes the seller's expected profit. Under the assumption that in small groups customers can learn each others' valuations, we show how Myerson's result can be improved to yield a higher payoff to the seller using a mechanism that offers groups of customers to buy bundles of items.

cs.GT↗

arXiv2012

A quantitative Gibbard-Satterthwaite theorem without neutrality

Recently, quantitative versions of the Gibbard-Satterthwaite theorem were proven for $k=3$ alternatives by Friedgut, Kalai, Keller and Nisan and for neutral functions on $k \geq 4$ alternatives by Isaksson, Kindler and Mossel. We prove a quantitative version of the Gibbard-Satterthwaite theorem for general social choice functions for any number $k \geq 3$ of alternatives. In particular we show that for a social choice function $f$ on $k \geq 3$ alternatives and $n$ voters, which is $ε$-far from the family of nonmanipulable functions, a uniformly chosen voter profile is manipulable with probability at least inverse polynomial in $n$, $k$, and $ε^{-1}$. Removing the neutrality assumption of previous theorems is important for multiple reasons. For one, it is known that there is a conflict between anonymity and neutrality, and since most common voting rules are anonymous, they cannot always be neutral. Second, virtual elections are used in many applications in artificial intelligence, where there are often restrictions on the outcome of the election, and so neutrality is not a natural assumption in these situations. Ours is a unified proof which in particular covers all previous cases established before. The proof crucially uses reverse hypercontractivity in addition to several ideas from the two previous proofs. Much of the work is devoted to understanding functions of a single voter, and in particular we also prove a quantitative Gibbard-Satterthwaite theorem for one voter.

math.CO↗

arXiv2011

On extracting common random bits from correlated sources

Suppose Alice and Bob receive strings of unbiased independent but noisy bits from some random source. They wish to use their respective strings to extract a common sequence of random bits with high probability but without communicating. How many such bits can they extract? The trivial strategy of outputting the first $k$ bits yields an agreement probability of $(1 - \eps)^k < 2^{-1.44k\eps}$, where $\eps$ is the amount of noise. We show that no strategy can achieve agreement probability better than $2^{-k\eps/(1 - \eps)}$. On the other hand, we show that when $k \geq 10 + 2 (1 - \eps) / \eps$, there exists a strategy which achieves an agreement probability of $0.1 (k\eps)^{-1/2} \cdot 2^{-k\eps/(1 - \eps)}$.

cs.IT↗

arXiv2011

Robust estimation of latent tree graphical models: Inferring hidden states with inexact parameters

Latent tree graphical models are widely used in computational biology, signal and image processing, and network tomography. Here we design a new efficient, estimation procedure for latent tree models, including Gaussian and discrete, reversible models, that significantly improves on previous sample requirement bounds. Our techniques are based on a new hidden state estimator which is robust to inaccuracies in estimated parameters. More precisely, we prove that latent tree models can be estimated with high probability in the so-called Kesten-Stigum regime with $O(log^2 n)$ samples where $n$ is the number of nodes.

math.PR↗