arXiv subjects

He Sun

Publications and source records attributed to He Sun.

121 records · Page 7Linked to original sources

arXiv2017

Distributed Graph Clustering and Sparsification

Graph clustering is a fundamental computational problem with a number of applications in algorithm design, machine learning, data mining, and analysis of social networks. Over the past decades, researchers have proposed a number of algorithmic design methods for graph clustering. Most of these methods, however, are based on complicated spectral techniques or convex optimisation, and cannot be directly applied for clustering many networks that occur in practice, whose information is often collected on different sites. Designing a simple and distributed clustering algorithm is of great interest, and has wide applications for processing big datasets. In this paper we present a simple and distributed algorithm for graph clustering: for a wide class of graphs that are characterised by a strong cluster-structure, our algorithm finishes in a poly-logarithmic number of rounds, and recovers a partition of the graph close to optimal. One of the main components behind our algorithm is a sampling scheme that, given a dense graph as input, produces a sparse subgraph that provably preserves the cluster-structure of the input. Compared with previous sparsification algorithms that require Laplacian solvers or involve combinatorial constructions, this component is easy to implement in a distributed way and runs fast in practice.

cs.DS↗

arXiv2017

An SDP-Based Algorithm for Linear-Sized Spectral Sparsification

For any undirected and weighted graph $G=(V,E,w)$ with $n$ vertices and $m$ edges, we call a sparse subgraph $H$ of $G$, with proper reweighting of the edges, a $(1+\varepsilon)$-spectral sparsifier if \[ (1-\varepsilon)x^{\intercal}L_Gx\leq x^{\intercal} L_{H} x\leq (1+\varepsilon) x^{\intercal} L_Gx \] holds for any $x\in\mathbb{R}^n$, where $L_G$ and $L_{H}$ are the respective Laplacian matrices of $G$ and $H$. Noticing that $Ω(m)$ time is needed for any algorithm to construct a spectral sparsifier and a spectral sparsifier of $G$ requires $Ω(n)$ edges, a natural question is to investigate, for any constant $\varepsilon$, if a $(1+\varepsilon)$-spectral sparsifier of $G$ with $O(n)$ edges can be constructed in $\tilde{O}(m)$ time, where the $\tilde{O}$ notation suppresses polylogarithmic factors. All previous constructions on spectral sparsification require either super-linear number of edges or $m^{1+Ω(1)}$ time. In this work we answer this question affirmatively by presenting an algorithm that, for any undirected graph $G$ and $\varepsilon>0$, outputs a $(1+\varepsilon)$-spectral sparsifier of $G$ with $O(n/\varepsilon^2)$ edges in $\tilde{O}(m/\varepsilon^{O(1)})$ time. Our algorithm is based on three novel techniques: (1) a new potential function which is much easier to compute yet has similar guarantees as the potential functions used in previous references; (2) an efficient reduction from a two-sided spectral sparsifier to a one-sided spectral sparsifier; (3) constructing a one-sided spectral sparsifier by a semi-definite program.

cs.DS↗

arXiv2017

Communication-Optimal Distributed Clustering

Clustering large datasets is a fundamental problem with a number of applications in machine learning. Data is often collected on different sites and clustering needs to be performed in a distributed manner with low communication. We would like the quality of the clustering in the distributed setting to match that in the centralized setting for which all the data resides on a single site. In this work, we study both graph and geometric clustering problems in two distributed models: (1) a point-to-point model, and (2) a model with a broadcast channel. We give protocols in both models which we show are nearly optimal by proving almost matching communication lower bounds. Our work highlights the surprising power of a broadcast channel for clustering problems; roughly speaking, to spectrally cluster $n$ points or $n$ vertices in a graph distributed across $s$ servers, for a worst-case partitioning the communication complexity in a point-to-point model is $n \cdot s$, while in the broadcast model it is $n + s$. A similar phenomenon holds for the geometric setting as well. We implement our algorithms and demonstrate this phenomenon on real life datasets, showing that our algorithms are also very efficient in practice.

cs.DS↗

arXiv2017

Partitioning Well-Clustered Graphs: Spectral Clustering Works!

In this paper we study variants of the widely used spectral clustering that partitions a graph into k clusters by (1) embedding the vertices of a graph into a low-dimensional space using the bottom eigenvectors of the Laplacian matrix, and (2) grouping the embedded points into k clusters via k-means algorithms. We show that, for a wide class of graphs, spectral clustering gives a good approximation of the optimal clustering. While this approach was proposed in the early 1990s and has comprehensive applications, prior to our work similar results were known only for graphs generated from stochastic models. We also give a nearly-linear time algorithm for partitioning well-clustered graphs based on computing a matrix exponential and approximate nearest neighbor data structures.

cs.DS↗

arXiv2015

Constructing Linear-Sized Spectral Sparsification in Almost-Linear Time

We present the first almost-linear time algorithm for constructing linear-sized spectral sparsification for graphs. This improves all previous constructions of linear-sized spectral sparsification, which requires $Ω(n^2)$ time. A key ingredient in our algorithm is a novel combination of two techniques used in literature for constructing spectral sparsification: Random sampling by effective resistance, and adaptive constructions based on barrier functions.

cs.DS↗

arXiv2014

Tight Bounds for Randomized Load Balancing on Arbitrary Network Topologies

We consider the problem of balancing load items (tokens) in networks. Starting with an arbitrary load distribution, we allow nodes to exchange tokens with their neighbors in each round. The goal is to achieve a distribution where all nodes have nearly the same number of tokens. For the continuous case where tokens are arbitrarily divisible, most load balancing schemes correspond to Markov chains, whose convergence is fairly well-understood in terms of their spectral gap. However, in many applications, load items cannot be divided arbitrarily, and we need to deal with the discrete case where the load is composed of indivisible tokens. This discretization entails a non-linear behavior due to its rounding errors, which makes this analysis much harder than in the continuous case. We investigate several randomized protocols for different communication models in the discrete case. As our main result, we prove that for any regular network in the matching model, all nodes have the same load up to an additive constant in (asymptotically) the same number of rounds as required in the continuous case. This generalizes and tightens the previous best result, which only holds for expander graphs, and demonstrates that there is almost no difference between the discrete and continuous cases. Our results also provide a positive answer to the question of how well discrete load balancing can be approximated by (continuous) Markov chains, which has been posed by many researchers.

cs.DM↗

arXiv2014

Balls into bins via local search: cover time and maximum load

We study a natural process for allocating m balls into n bins that are organized as the vertices of an undirected graph G. Balls arrive one at a time. When a ball arrives, it first chooses a vertex u in G uniformly at random. Then the ball performs a local search in G starting from u until it reaches a vertex with local minimum load, where the ball is finally placed on. Then the next ball arrives and this procedure is repeated. For the case m = n, we give an upper bound for the maximum load on graphs with bounded degrees. We also propose the study of the cover time of this process, which is defined as the smallest m so that every bin has at least one ball allocated to it. We establish an upper bound for the cover time on graphs with bounded degrees. Our bounds for the maximum load and the cover time are tight when the graph is transitive or sufficiently homogeneous. We also give upper bounds for the maximum load when m > n.

math.PR↗

arXiv2013

Randomness-Efficient Rumor Spreading

We study the classical rumor spreading problem, which is used to spread information in an unknown network with $n$ nodes. We present the first protocol for any expander graph $G$ with $n$ nodes and minimum degree $Θ(n)$ such that, the protocol informs every node in $O(\log n)$ rounds with high probability, and uses $O(\log n\log\log n)$ random bits in total. The runtime of our protocol is tight, and the randomness requirement of $O(\log n\log\log n)$ random bits almost matches the lower bound of $Ω(\log n)$ random bits. We further study rumor spreading protocols for more general graphs, and for several graph topologies our protocols are as fast as the classical protocol and use $\tilde{O}(\log n)$ random bits in total, in contrast to $O(n\log^2n)$ random bits used in the well-known rumor spreading push protocol. These results together give us almost full understanding of the randomness requirement for this basic epidemic process. Our protocols rely on a novel reduction between rumor spreading processes and branching programs, and this reduction provides a general framework to derandomize these complex and distributed epidemic processes. Interestingly, one cannot simply apply PRGs for branching programs as rumor spreading process is not characterized by small-space computation. Our protocols require the composition of several pseudorandom objects, e.g. pseudorandom generators, and pairwise independent generators. Besides designing rumor spreading protocols, the techniques developed here may have applications in studying the randomness complexity of distributed algorithms.

cs.DS↗

arXiv2013

Gossip vs. Markov Chains, and Randomness-Efficient Rumor Spreading

We study gossip algorithms for the rumor spreading problem which asks one node to deliver a rumor to all nodes in an unknown network. We present the first protocol for any expander graph $G$ with $n$ nodes such that, the protocol informs every node in $O(\log n)$ rounds with high probability, and uses $\tilde{O}(\log n)$ random bits in total. The runtime of our protocol is tight, and the randomness requirement of $\tilde{O}(\log n)$ random bits almost matches the lower bound of $Ω(\log n)$ random bits for dense graphs. We further show that, for many graph families, polylogarithmic number of random bits in total suffice to spread the rumor in $O(\mathrm{poly}\log n)$ rounds. These results together give us an almost complete understanding of the randomness requirement of this fundamental gossip process. Our analysis relies on unexpectedly tight connections among gossip processes, Markov chains, and branching programs. First, we establish a connection between rumor spreading processes and Markov chains, which is used to approximate the rumor spreading time by the mixing time of Markov chains. Second, we show a reduction from rumor spreading processes to branching programs, and this reduction provides a general framework to derandomize gossip processes. In addition to designing rumor spreading protocols, these novel techniques may have applications in studying parallel and multiple random walks, and randomness complexity of distributed algorithms.

cs.DC↗

arXiv2013

Counting Hypergraphs in Data Streams

We present the first streaming algorithm for counting an arbitrary hypergraph $H$ of constant size in a massive hypergraph $G$. Our algorithm can handle both edge-insertions and edge-deletions, and is applicable for the distributed setting. Moreover, our approach provides the first family of graph polynomials for the hypergraph counting problem. Because of the close relationship between hypergraphs and set systems, our approach may have applications in studying similar problems.

cs.DS↗

arXiv2012

Geometric Pricing: How Low Dimensionality Helps in Approximability

Consider the following toy problem. There are $m$ rectangles and $n$ points on the plane. Each rectangle $R$ is a consumer with budget $B_R$, who is interested in purchasing the cheapest item (point) inside R, given that she has enough budget. Our job is to price the items to maximize the revenue. This problem can also be defined on higher dimensions. We call this problem the geometric pricing problem. In this paper, we study a new class of problems arising from a geometric aspect of the pricing problem. It intuitively captures typical real-world assumptions that have been widely studied in marketing research, healthcare economics, etc. It also helps classify other well-known pricing problems, such as the highway pricing problem and the graph vertex pricing problem on planar and bipartite graphs. Moreover, this problem turns out to have close connections to other natural geometric problems such as the geometric versions of the unique coverage and maximum feasible subsystem problems. We show that the low dimensionality arising in this pricing problem does lead to improved approximation ratios, by presenting sublinear-approximation algorithms for two central versions of the problem: unit-demand uniform-budget min-buying and single-minded pricing problems. Our algorithm is obtained by combining algorithmic pricing and geometric techniques. These results suggest that considering geometric aspect might be a promising research direction in obtaining improved approximation algorithms for such pricing problems. To the best of our knowledge, this is one of very few problems in the intersection between geometry and algorithmic pricing areas. Thus its study may lead to new algorithmic techniques that could benefit both areas.

cs.GT↗

arXiv2012

Balls into Bins via Local Search

We propose a natural process for allocating n balls into n bins that are organized as the vertices of an undirected graph G. Each ball first chooses a vertex u in G uniformly at random. Then the ball performs a local search in G starting from u until it reaches a vertex with local minimum load, where the ball is finally placed on. In our main result, we prove that this process yields a maximum load of only Θ(\log \log n) on expander graphs. In addition, we show that for d-dimensional grids the maximum load is Θ\Big(\big(\frac{\log n}{\log \log n}\big)^{\frac{1}{d+1}}\Big). Finally, for almost regular graphs with minimum degree Ω(\log n), we prove that the maximum load is constant and also reveal a fundamental difference between random and arbitrary tie-breaking rules.

math.PR↗

arXiv2012

Deterministic Polynomial-Time Algorithms for Designing Short DNA Words

Designing short DNA words is a problem of constructing a set (i.e., code) of n DNA strings (i.e., words) with the minimum length such that the Hamming distance between each pair of words is at least k and the n words satisfy a set of additional constraints. This problem has applications in, e.g., DNA self-assembly and DNA arrays. Previous works include those that extended results from coding theory to obtain bounds on code and word sizes for biologically motivated constraints and those that applied heuristic local searches, genetic algorithms, and randomized algorithms. In particular, Kao, Sanghi, and Schweller (2009) developed polynomial-time randomized algorithms to construct n DNA words of length within a multiplicative constant of the smallest possible word length (e.g., 9 max{log n, k}) that satisfy various sets of constraints with high probability. In this paper, we give deterministic polynomial-time algorithms to construct DNA words based on derandomization techniques. Our algorithms can construct n DNA words of shorter length (e.g., 2.1 log n + 6.28 k) and satisfy the same sets of constraints as the words constructed by the algorithms of Kao et al. Furthermore, we extend these new algorithms to construct words that satisfy a larger set of constraints for which the algorithms of Kao et al. do not work.

cs.DS↗