arXiv subjects

Luc Devroye

Publications and source records attributed to Luc Devroye.

At least 73 records · Page 4Linked to original sources

arXiv2015

The Analysis of Kademlia for random IDs

Kademlia is the de facto standard searching algorithm for P2P (peer-to-peer) networks on the Internet. In our earlier work, we introduced two slightly different models for Kademlia and studied how many steps it takes to search for a target node by using Kademlia's searching algorithm. The first model, in which nodes of the network are labelled with deterministic IDs, had been discussed in that paper. The second one, in which nodes are labelled with random IDs, which we call the Random ID Model, was only briefly mentioned. Refined results with detailed proofs for this model are given in this paper. Our analysis shows that with high probability it takes about $c \log n$ steps to locate any node, where $n$ is the total number of nodes in the network and $c$ is a constant that does not depend on $n$.

cs.DS↗

arXiv2014

Explosion and linear transit times in infinite trees

Let $T$ be an infinite rooted tree with weights $w_e$ assigned to its edges. Denote by $m_n(T)$ the minimum weight of a path from the root to a node of the $n$th generation. We consider the possible behaviour of $m_n(T)$ with focus on the two following cases: we say $T$ is explosive if \[ \lim_{n\to \infty}m_n(T) < \infty, \] and say that $T$ exhibits linear growth if \[ \liminf_{n\to \infty} \frac{m_n(T)}{n} > 0. \] We consider a class of infinite randomly weighted trees related to the Poisson-weighted infinite tree, and determine precisely which trees in this class have linear growth almost surely. We then apply this characterization to obtain new results concerning the event of explosion in infinite randomly weighted spherically-symmetric trees, answering a question of Pemantle and Peres. As a further application, we consider the random real tree generated by attaching sticks of deterministic decreasing lengths, and determine for which sequences of lengths the tree has finite height almost surely.

math.PR↗

arXiv2014

Almost optimal sparsification of random geometric graphs

A random geometric irrigation graph $Γ_n(r_n,ξ)$ has $n$ vertices identified by $n$ independent uniformly distributed points $X_1,\ldots,X_n$ in the unit square $[0,1]^2$. Each point $X_i$ selects $ξ_i$ neighbors at random, without replacement, among those points $X_j$ ($j\neq i$) for which $\|X_i-X_j\| < r_n$, and the selected vertices are connected to $X_i$ by an edge. The number $ξ_i$ of the neighbors is an integer-valued random variable, chosen independently with identical distribution for each $X_i$ such that $ξ_i$ satisfies $1\le ξ_i \le κ$ for a constant $κ>1$. We prove that when $r_n = γ_n \sqrt{\log n/n}$ for $γ_n \to \infty$ with $γ_n =o(n^{1/6}/\log^{5/6}n)$, then the random geometric irrigation graph experiences explosive percolation in the sense that when $\mathbf E ξ_i=1$, then the largest connected component has size $o(n)$ but if $\mathbf E ξ_i >1$, then the size of the largest connected component is with high probability $n-o(n)$. This offers a natural non-centralized sparsification of a random geometric graph that is mostly connected.

math.PR↗

arXiv2014

Connectivity of sparse Bluetooth networks

Consider a random geometric graph defined on $n$ vertices uniformly distributed in the $d$-dimensional unit torus. Two vertices are connected if their distance is less than a "visibility radius" $r_n$. We consider {\sl Bluetooth networks} that are locally sparsified random geometric graphs. Each vertex selects $c$ of its neighbors in the random geometric graph at random and connects only to the selected points. We show that if the visibility radius is at least of the order of $n^{-(1-δ)/d}$ for some $δ> 0$, then a constant value of $c$ is sufficient for the graph to be connected, with high probability. It suffices to take $c \ge \sqrt{(1+ε)/δ} + K$ for any positive $ε$ where $K$ is a constant depending on $d$ only. On the other hand, with $c\le \sqrt{(1-ε)/δ}$, the graph is disconnected, with high probability.

math.PR↗

arXiv2013

Protected nodes and fringe subtrees in some random trees

We study protected nodes in various classes of random rooted trees by putting them in the general context of fringe subtrees introduced by Aldous (1991). Several types of random trees are considered: simply generated trees (or conditioned Galton-Watson trees), which includes several cases treated separately by other authors, binary search trees and random recursive trees. This gives unified and simple proofs of several earlier results, as well as new results.

math.PR↗

arXiv2013

A Probabilistic Analysis of Kademlia Networks

Kademlia is currently the most widely used searching algorithm in P2P (peer-to-peer) networks. This work studies an essential question about Kademlia from a mathematical perspective: how long does it take to locate a node in the network? To answer it, we introduce a random graph K and study how many steps are needed to locate a given vertex in K using Kademlia's algorithm, which we call the routing time. Two slightly different versions of K are studied. In the first one, vertices of K are labelled with fixed IDs. In the second one, vertices are assumed to have randomly selected IDs. In both cases, we show that the routing time is about c*log(n), where n is the number of nodes in the network and c is an explicitly described constant.

cs.NI↗

arXiv2013

Cellular Tree Classifiers

The cellular tree classifier model addresses a fundamental problem in the design of classifiers for a parallel or distributed computing world: Given a data set, is it sufficient to apply a majority rule for classification, or shall one split the data into two or more parts and send each part to a potentially different computer (or cell) for further processing? At first sight, it seems impossible to define with this paradigm a consistent classifier as no cell knows the "original data size", $n$. However, we show that this is not so by exhibiting two different consistent classifiers. The consistency is universal but is only shown for distributions with nonatomic marginals.

stat.ML↗

arXiv2013

On explosions in heavy-tailed branching random walks

Consider a branching random walk on $\mathbb{R}$, with offspring distribution Z and nonnegative displacement distribution W. We say that explosion occurs if an infinite number of particles may be found within a finite distance of the origin. In this paper, we investigate this phenomenon when the offspring distribution Z is heavy-tailed. Under an appropriate condition, we are able to characterize the pairs (Z, W) for which explosion occurs, by demonstrating the equivalence of explosion with a seemingly much weaker event: that the sum over generations of the minimum displacement in each generation is finite. Furthermore, we demonstrate that our condition on the tail is best possible for this equivalence to occur. We also investigate, under additional smoothness assumptions, the behavior of $M_n$, the position of the particle in generation n closest to the origin, when explosion does not occur (and hence $\lim_{n\rightarrow\infty}M_n=\infty$).

math.PR↗

arXiv2013

Prediction by Random-Walk Perturbation

We propose a version of the follow-the-perturbed-leader online prediction algorithm in which the cumulative losses are perturbed by independent symmetric random walks. The forecaster is shown to achieve an expected regret of the optimal order O(sqrt(n log N)) where n is the time horizon and N is the number of experts. More importantly, it is shown that the forecaster changes its prediction at most O(sqrt(n log N)) times, in expectation. We also extend the analysis to online combinatorial optimization and show that even in this more general setting, the forecaster rarely switches between experts while having a regret of near-optimal order.

cs.LG↗

arXiv2012

Depth properties of scaled attachment random recursive trees

We study depth properties of a general class of random recursive trees where each node i attaches to the random node iX_i and X_0, ..., X_n is a sequence of i.i.d. random variables taking values in [0,1). We call such trees scaled attachment random recursive trees (SARRT). We prove that the typical depth D_n, the maximum depth (or height) H_n and the minimum depth M_n of a SARRT are asymptotically given by D_n \sim μ^{-1} \log n, H_n \sim α_{\max} \log n and M_n \sim α_{\min} \log n where μ, α_{\max} and α_{\min} are constants depending only on the distribution of X_0 whenever X_0 has a density. In particular, this gives a new elementary proof for the height of uniform random recursive trees H_n \sim e \log n that does not use branching random walks.

math.PR↗

arXiv2012

Connectivity of inhomogeneous random graphs

We find conditions for the connectivity of inhomogeneous random graphs with intermediate density. Our results generalize the classical result for G(n, p), when p = c log n/n. We draw n independent points X_i from a general distribution on a separable metric space, and let their indices form the vertex set of a graph. An edge (i,j) is added with probability min(1, \K(X_i,X_j) log n/n), where \K \ge 0 is a fixed kernel. We show that, under reasonably weak assumptions, the connectivity threshold of the model can be determined.

math.PR↗

arXiv2012

A Note on Interference in Random Point Sets

The (maximum receiver-centric) interference of a geometric graph (von Rickenbach etal (2005)) is studied. It is shown that, with high probability, the following results hold for a set, V, of n points independently and uniformly distributed in the unit d-cube, for constant dimension d: (1) there exists a connected graph with vertex set V that has interference O((log n)^{1/3}); (2) no connected graph with vertex set V has interference o((log n)^{1/4}); and (3) the minimum spanning tree of $V$ has interference Theta((\log n)^{1/2}).

cs.CG↗

arXiv2012

An Affine Invariant $k$-Nearest Neighbor Regression Estimate

We design a data-dependent metric in $\mathbb R^d$ and use it to define the $k$-nearest neighbors of a given point. Our metric is invariant under all affine transformations. We show that, with this metric, the standard $k$-nearest neighbor regression estimate is asymptotically consistent under the usual conditions on $k$, and minimal requirements on the input data.

math.ST↗

arXiv2011

Random hyperplane search trees in high dimensions

Given a set S of n \geq d points in general position in R^d, a random hyperplane split is obtained by sampling d points uniformly at random without replacement from S and splitting based on their affine hull. A random hyperplane search tree is a binary space partition tree obtained by recursive application of random hyperplane splits. We investigate the structural distributions of such random trees with a particular focus on the growth with d. A blessing of dimensionality arises--as d increases, random hyperplane splits more closely resemble perfectly balanced splits; in turn, random hyperplane search trees more closely resemble perfectly balanced binary search trees. We prove that, for any fixed dimension d, a random hyperplane search tree storing n points has height at most (1 + O(1/sqrt(d))) log_2 n and average element depth at most (1 + O(1/d)) log_2 n with high probability as n \rightarrow \infty. Further, we show that these bounds are asymptotically optimal with respect to d.

cs.CG↗

arXiv2011

Connectivity threshold for Bluetooth graphs

We study the connectivity properties of random Bluetooth graphs that model certain "ad hoc" wireless networks. The graphs are obtained as "irrigation subgraphs" of the well-known random geometric graph model. There are two parameters that control the model: the radius $r$ that determines the "visible neighbors" of each node and the number of edges $c$ that each node is allowed to send to these. The randomness comes from the underlying distribution of data points in space and from the choices of each vertex. We prove that no connectivity can take place with high probability for a range of parameters $r, c$ and completely characterize the connectivity threshold (in $c$) for values of $r$ close the critical value for connectivity in the underlying random geometric graph.

math.PR↗

arXiv2010

On combinatorial testing problems

We study a class of hypothesis testing problems in which, upon observing the realization of an $n$-dimensional Gaussian vector, one has to decide whether the vector was drawn from a standard normal distribution or, alternatively, whether there is a subset of the components belonging to a certain given class of sets whose elements have been ``contaminated,'' that is, have a mean different from zero. We establish some general conditions under which testing is possible and others under which testing is hopeless with a small risk. The combinatorial and geometric structure of the class of sets is shown to play a crucial role. The bounds are illustrated on various examples.

math.ST↗

arXiv2010

Sub-Gaussian tail bounds for the width and height of conditioned Galton--Watson trees

We study the height and width of a Galton--Watson tree with offspring distribution B satisfying E(B)=1, 0 < Var(B) < infinity, conditioned on having exactly n nodes. Under this conditioning, we derive sub-Gaussian tail bounds for both the width (largest number of nodes in any level) and height (greatest level containing a node); the bounds are optimal up to constant factors in the exponent. Under the same conditioning, we also derive essentially optimal upper tail bounds for the number of nodes at level k, for 1 <= k <= n.

math.PR↗

arXiv2010

The dilation of the Delaunay triangulation is greater than π/2

Consider the Delaunay triangulation T of a set P of points in the plane as a Euclidean graph, in which the weight of every edge is its length. It has long been conjectured that the dilation in T of any pair p, p \in P, which is the ratio of the length of the shortest path from p to p' in T over the Euclidean distance ||pp'||, can be at most π/2 \approx 1.5708. In this paper, we show how to construct point sets in convex position with dilation > 1.5810 and in general position with dilation > 1.5846. Furthermore, we show that a sufficiently large set of points drawn independently from any distribution will in the limit approach the worst-case dilation for that distribution.

cs.CG↗