arXiv subjects

Henning Meyerhenke

Publications and source records attributed to Henning Meyerhenke.

At least 37 records · Page 2Linked to original sources

arXiv2021

Fully-dynamic Weighted Matching Approximation in Practice

Finding large or heavy matchings in graphs is a ubiquitous combinatorial optimization problem. In this paper, we engineer the first non-trivial implementations for approximating the dynamic weighted matching problem. Our first algorithm is based on random walks/paths combined with dynamic programming. The second algorithm has been introduced by Stubbs and Williams without an implementation. Roughly speaking, their algorithm uses dynamic unweighted matching algorithms as a subroutine (within a multilevel approach); this allows us to use previous work on dynamic unweighted matching algorithms as a black box in order to obtain a fully-dynamic weighted matching algorithm. We empirically study the algorithms on an extensive set of dynamic instances and compare them with optimal weighted matchings. Our experiments show that the random walk algorithm typically fares much better than Stubbs/Williams (regarding the time/quality tradeoff), and its results are often not far from the optimum.

cs.DS↗

arXiv2021

Approximation of the Diagonal of a Laplacian's Pseudoinverse for Complex Network Analysis

The ubiquity of massive graph data sets in numerous applications requires fast algorithms for extracting knowledge from these data. We are motivated here by three electrical measures for the analysis of large small-world graphs $G = (V, E)$ -- i.e., graphs with diameter in $O(\log |V|)$, which are abundant in complex network analysis. From a computational point of view, the three measures have in common that their crucial component is the diagonal of the graph Laplacian's pseudoinverse, $L^\dagger$. Computing diag$(L^\dagger)$ exactly by pseudoinversion, however, is as expensive as dense matrix multiplication -- and the standard tools in practice even require cubic time. Moreover, the pseudoinverse requires quadratic space -- hardly feasible for large graphs. Resorting to approximation by, e.g., using the Johnson-Lindenstrauss transform, requires the solution of $O(\log |V| / ε^2)$ Laplacian linear systems to guarantee a relative error, which is still very expensive for large inputs. In this paper, we present a novel approximation algorithm that requires the solution of only one Laplacian linear system. The remaining parts are purely combinatorial -- mainly sampling uniform spanning trees, which we relate to diag$(L^\dagger)$ via effective resistances. For small-world networks, our algorithm obtains a $\pm ε$-approximation with high probability, in a time that is nearly-linear in $|E|$ and quadratic in $1 / ε$. Another positive aspect of our algorithm is its parallel nature due to independent sampling. We thus provide two parallel implementations of our algorithm: one using OpenMP, one MPI + OpenMP. In our experiments against the state of the art, our algorithm (i) yields more accurate results, (ii) is much faster and more memory-efficient, and (iii) obtains good parallel speedups, in particular in the distributed setting.

cs.DS↗

arXiv2021

New Approximation Algorithms for Forest Closeness Centrality -- for Individual Vertices and Vertex Groups

The emergence of massive graph data sets requires fast mining algorithms. Centrality measures to identify important vertices belong to the most popular analysis methods in graph mining. A measure that is gaining attention is forest closeness centrality; it is closely related to electrical measures using current flow but can also handle disconnected graphs. Recently, [Jin et al., ICDM'19] proposed an algorithm to approximate this measure probabilistically. Their algorithm processes small inputs quickly, but does not scale well beyond hundreds of thousands of vertices. In this paper, we first propose a different approximation algorithm; it is up to two orders of magnitude faster and more accurate in practice. Our method exploits the strong connection between uniform spanning trees and forest distances by adapting and extending recent approximation algorithms for related single-vertex problems. This results in a nearly-linear time algorithm with an absolute probabilistic error guarantee. In addition, we are the first to consider the problem of finding an optimal group of vertices w.r.t. forest closeness. We prove that this latter problem is NP-hard; to approximate it, we adapt a greedy algorithm by [Li et al., WWW'19], which is based on (partial) matrix inversion. Moreover, our experiments show that on disconnected graphs, group forest closeness outperforms existing centrality measures in the context of semi-supervised vertex classification.

cs.DS↗

arXiv2020

Systematic partitioning of proteins for quantum-chemical fragmentation methods using graph algorithms

Quantum-chemical fragmentation methods offer an efficient approach for the treatment of large proteins, in particular if local target quantities such as protein--ligand interaction energies, enzymatic reaction energies, or spectroscopic properties of embedded chromophores are sought. However, the accuracy that is achievable for such local target quantities intricately depends on how the protein is partitioned into smaller fragments. While the commonly employed naïve approach of using fragments with a fixed size is widely used, it can result in large and unpredictable errors when varying the fragment size. Here, we present a systematic partitioning scheme that aims at minimizing the fragmentation error of a local target quantity for a given maximum fragment size. To this end, we construct a weighted graph representation of the protein, in which the amino acids constitute the nodes. These nodes are connected by edges weighted with an estimate for the fragmentation error that is expected when cutting this edge. This allows us to employ graph partitioning algorithms provided by computer science to determine near-optimal partitions of the protein. We apply this scheme to a test set of six proteins representing various prototypical applications of quantum-chemical fragmentation methods using a simplified molecular fractionation with conjugate caps (MFCC) approach with hydrogen caps. We show that our graph-based scheme consistently improves upon the naïve approach.

physics.chem-ph↗

arXiv2020

Distributing Sparse Matrix/Graph Applications in Heterogeneous Clusters -- an Experimental Study

Many problems in scientific and engineering applications contain sparse matrices or graphs as main input objects, e.g. numerical simulations on meshes. Large inputs are abundant these days and require parallel processing for memory size and speed. To optimize the execution of such simulations on cluster systems, the input problem needs to be distributed suitably onto the processing units (PUs). More and more frequently, such clusters contain different CPUs or a combination of CPUs and GPUs. This heterogeneity makes the load distribution problem quite challenging. Our study is motivated by the observation that established partitioning tools do not handle such heterogeneous distribution problems as well as homogeneous ones. In this paper, we first formulate the problem of balanced load distribution for heterogeneous architectures as a multi-objective, single-constraint optimization problem. We then split the problem into two phases and propose a greedy approach to determine optimal block sizes for each PU. These block sizes are then fed into numerous existing graph partitioners, for us to examine how well they handle the above problem. One of the tools we consider is an extension of our own previous work (von Looz et al, ICPP'18) called Geographer. Our experiments on well-known benchmark meshes indicate that only two tools under consideration are able to yield good quality. These two are Parmetis (both the geometric and the combinatorial variant) and Geographer. While Parmetis is faster, Geographer yields better quality on average.

cs.DC↗

arXiv2020

Estimating Current-Flow Closeness Centrality with a Multigrid Laplacian Solver

Matrices associated with graphs, such as the Laplacian, lead to numerous interesting graph problems expressed as linear systems. One field where Laplacian linear systems play a role is network analysis, e. g. for certain centrality measures that indicate if a node (or an edge) is important in the network. One such centrality measure is current-flow closeness. To allow network analysis workflows to profit from a fast Laplacian solver, we provide an implementation of the LAMG multigrid solver in the NetworKit package, facilitating the computation of current-flow closeness values or related quantities. Our main contribution consists of two algorithms that accelerate the current-flow computation for one node or a reasonably small node subset significantly. One sampling-based algorithm provides an unbiased estimation of the related electrical farness, the other one is based on the Johnson-Lindenstrauss transform. Our inexact algorithms lead to very accurate results in practice. Thanks to them one is now able to compute an estimation of current-flow closeness of one node on networks with tens of millions of nodes and edges within seconds or a few minutes. From a network analytical point of view, our experiments indicate that current-flow closeness can discriminate among different nodes significantly better than traditional shortest-path closeness and is also considerably more resistant to noise -- we thus show that two known drawbacks of shortest-path closeness are alleviated by the current-flow variant.

cs.DS↗

arXiv2020

Group-Harmonic and Group-Closeness Maximization -- Approximation and Engineering

Centrality measures characterize important nodes in networks. Efficiently computing such nodes has received a lot of attention. When considering the generalization of computing central groups of nodes, challenging optimization problems occur. In this work, we study two such problems, group-harmonic maximization and group-closeness maximization both from a theoretical and from an algorithm engineering perspective. On the theoretical side, we obtain the following results. For group-harmonic maximization, unless $P=NP$, there is no polynomial-time algorithm that achieves an approximation factor better than $1-1/e$ (directed) and $1-1/(4e)$ (undirected), even for unweighted graphs. On the positive side, we show that a greedy algorithm achieves an approximation factor of $λ(1-2/e)$ (directed) and $λ(1-1/e)/2$ (undirected), where $λ$ is the ratio of minimal and maximal edge weights. For group-closeness maximization, the undirected case is $NP$-hard to be approximated to within a factor better than $1-1/(e+1)$ and a constant approximation factor is achieved by a local-search algorithm. For the directed case, however, we show that, for any $ε<1/2$, the problem is $NP$-hard to be approximated within a factor of $4|V|^{-ε}$. From the algorithm engineering perspective, we provide efficient implementations of the above greedy and local search algorithms. In our experimental study we show that, on small instances where an optimum solution can be computed in reasonable time, the quality of both the greedy and the local search algorithms come very close to the optimum. On larger instances, our local search algorithms yield results with superior quality compared to existing greedy and local search solutions, at the cost of additional running time. We thus advocate local search for scenarios where solution quality is of highest concern.

cs.DS↗

arXiv2020

Combined Centrality Measures for an Improved Characterization of Influence Spread in Social Networks

Influence Maximization (IM) aims at finding the most influential users in a social network, i. e., users who maximize the spread of an opinion within a certain propagation model. Previous work investigated the correlation between influence spread and nodal centrality measures to bypass more expensive IM simulations. The results were promising but incomplete, since these studies investigated the performance (i. e., the ability to identify influential users) of centrality measures only in restricted settings, e. g., in undirected/unweighted networks and/or within a propagation model less common for IM. In this paper, we first show that good results within the Susceptible- Infected-Removed (SIR) propagation model for unweighted and undirected networks do not necessarily transfer to directed or weighted networks under the popular Independent Cascade (IC) propagation model. Then, we identify a set of centrality measures with good performance for weighted and directed networks within the IC model. Our main contribution is a new way to combine the centrality measures in a closed formula to yield even better results. Additionally, we also extend gravitational centrality (GC) with the proposed combined centrality measures. Our experiments on 50 real-world data sets show that our proposed centrality measures outperform well-known centrality measures and the state-of-the art GC measure significantly. social networks, influence maximization, centrality measures, IC propagation model, influential spreaders

cs.SI↗

arXiv2020

High-Quality Hierarchical Process Mapping

Partitioning graphs into blocks of roughly equal size such that few edges run between blocks is a frequently needed operation when processing graphs on a parallel computer. When a topology of a distributed system is known an important task is then to map the blocks of the partition onto the processors such that the overall communication cost is reduced. We present novel multilevel algorithms that integrate graph partitioning and process mapping. Important ingredients of our algorithm include fast label propagation, more localized local search, initial partitioning, as well as a compressed data structure to compute processor distances without storing a distance matrix. Experiments indicate that our algorithms speed up the overall mapping process and, due to the integrated multilevel approach, also find much better solutions in practice. For example, one configuration of our algorithm yields better solutions than the previous state-of-the-art in terms of mapping quality while being a factor 62 faster. Compared to the currently fastest iterated multilevel mapping algorithm Scotch, we obtain 16% better solutions while investing slightly more running time.

cs.DS↗

arXiv2019

Local Search for Group Closeness Maximization on Big Graphs

In network analysis and graph mining, closeness centrality is a popular measure to infer the importance of a vertex. Computing closeness efficiently for individual vertices received considerable attention. The NP-hard problem of group closeness maximization, in turn, is more challenging: the objective is to find a vertex group that is central as a whole and state-of-the-art heuristics for it do not scale to very big graphs yet. In this paper, we present new local search heuristics for group closeness maximization. By using randomized approximation techniques and dynamic data structures, our algorithms are often able to perform locally optimal decisions efficiently. The final result is a group with high (but not optimal) closeness centrality. We compare our algorithms to the current state-of-the-art greedy heuristic both on weighted and on unweighted real-world graphs. For graphs with hundreds of millions of edges, our local search algorithms take only around ten minutes, while greedy requires more than ten hours. Overall, our new algorithms are between one and two orders of magnitude faster, depending on the desired group size and solution quality. For example, on weighted graphs and $k = 10$, our algorithms yield solutions of $12,4\%$ higher quality, while also being $793,6\times$ faster. For unweighted graphs and $k = 10$, we achieve solutions within $99,4\%$ of the state-of-the-art quality while being $127,8\times$ faster.

cs.DS↗

arXiv2019

Group Centrality Maximization for Large-scale Graphs

The study of vertex centrality measures is a key aspect of network analysis. Naturally, such centrality measures have been generalized to groups of vertices; for popular measures it was shown that the problem of finding the most central group is $\mathcal{NP}$-hard. As a result, approximation algorithms to maximize group centralities were introduced recently. Despite a nearly-linear running time, approximation algorithms for group betweenness and (to a lesser extent) group closeness are rather slow on large networks due to high constant overheads. That is why we introduce GED-Walk centrality, a new submodular group centrality measure inspired by Katz centrality. In contrast to closeness and betweenness, it considers walks of any length rather than shortest paths, with shorter walks having a higher contribution. We define algorithms that (i) efficiently approximate the GED-Walk score of a given group and (ii) efficiently approximate the (proved to be $\mathcal{NP}$-hard) problem of finding a group with highest GED-Walk score. Experiments on several real-world datasets show that scores obtained by GED-Walk improve performance on common graph mining tasks such as collective classification and graph-level classification. An evaluation of empirical running times demonstrates that maximizing GED-Walk (in approximation) is two orders of magnitude faster compared to group betweenness approximation and for group sizes $\leq 100$ one to two orders faster than group closeness approximation. For graphs with tens of millions of edges, approximate GED-Walk maximization typically needs less than one minute. Furthermore, our experiments suggest that the maximization algorithms scale linearly with the size of the input graph and the size of the group.

cs.DS↗

arXiv2019

Scaling Betweenness Approximation to Billions of Edges by MPI-based Adaptive Sampling

Betweenness centrality is one of the most popular vertex centrality measures in network analysis. Hence, many (sequential and parallel) algorithms to compute or approximate betweenness have been devised. Recent algorithmic advances have made it possible to approximate betweenness very efficiently on shared-memory architectures. Yet, the best shared-memory algorithms can still take hours of running time for large graphs, especially for graphs with a high diameter or when a small relative error is required. In this work, we present an MPI-based generalization of the state-of-the-art shared-memory algorithm for betweenness approximation. This algorithm is based on adaptive sampling; our parallelization strategy can be applied in the same manner to adaptive sampling algorithms for other problems. In experiments on a 16-node cluster, our MPI-based implementation is by a factor of 16.1x faster than the state-of-the-art shared-memory implementation when considering our parallelization focus -- the adaptive sampling phase -- only. For the complete algorithm, we obtain an average (geom. mean) speedup factor of 7.4x over the state of the art. For some previously very challenging inputs, this speedup is much higher. As a result, our algorithm is the first to approximate betweenness centrality on graphs with several billion edges in less than ten minutes with high accuracy.

cs.DC↗

arXiv2019

Scaling up Group Closeness Maximization

Closeness is a widely-used centrality measure in social network analysis. For a node it indicates the reciprocal of the average shortest-path distance to the other nodes of the network. While the identification of the k nodes with highest closeness received significant attention, many applications are actually interested in finding a group of nodes that is central as a whole. For this problem, only recently a greedy algorithm has been proposed [Chen et al., ADC 2016]. The approximation factor of (1 - 1/e) proposed by Chen et al. for this algorithm does not hold, though, as we show in this version of our paper. Since their implementation of the greedy algorithm was still too slow for large networks, Chen et al. also proposed a heuristic without approximation guarantee. In the present paper we develop new techniques to speed up the greedy algorithm. Compared to the previous implementation, our approach is orders of magnitude faster and, compared to the heuristic proposed by Chen et al., we always find a solution with better quality in a comparable running time in our experiments. Our method Greedy++ allows us to estimate the group with maximum closeness on networks with up to hundreds of millions of edges in minutes or at most a few hours. The greedy approach by [Chen et al., ADC 2016] would take several days already on networks with hundreds of thousands of edges. Our experiments show that the solution found by Greedy++ is actually very close to the optimum (...) Note: This paper version fixes the issue of relying on the presumed (but incorrect) submodularity of group closeness. While this has implications on the theoretical assessment of the greedy algorithm, our algorithm variant and its implementation remain unaffected. The reason is that Greedy++ relies (among others) on the supermodularity of farness, which does hold.

cs.DS↗

arXiv2019

Guidelines for Experimental Algorithmics in Network Analysis

The field of network science is a highly interdisciplinary area; for the empirical analysis of network data, it draws algorithmic methodologies from several research fields. Hence, research procedures and descriptions of the technical results often differ, sometimes widely. In this paper we focus on methodologies for the experimental part of algorithm engineering for network analysis -- an important ingredient for a research area with empirical focus. More precisely, we unify and adapt existing recommendations from different fields and propose universal guidelines -- including statistical analyses -- for the systematic evaluation of network analysis algorithms. This way, the behavior of newly proposed algorithms can be properly assessed and comparisons to existing solutions become meaningful. Moreover, as the main technical contribution, we provide SimexPal, a highly automated tool to perform and analyze experiments following our guidelines. To illustrate the merits of SimexPal and our guidelines, we apply them in a case study: we design, perform, visualize and evaluate experiments of a recent algorithm for approximating betweenness centrality, an important problem in network analysis. In summary, both our guidelines and SimexPal shall modernize and complement previous efforts in experimental algorithmics; they are not only useful for network analysis, but also in related contexts.

cs.SI↗

arXiv2019

Parallel Adaptive Sampling with almost no Synchronization

Approximation via sampling is a widespread technique whenever exact solutions are too expensive. In this paper, we present techniques for an efficient parallelization of adaptive (a. k. a. progressive) sampling algorithms on multi-threaded shared-memory machines. Our basic algorithmic technique requires no synchronization except for atomic load-acquire and store-release operations. It does, however, require O(n) memory per thread, where n is the size of the sampling state. We present variants of the algorithm that either reduce this memory consumption to O(1) or ensure that deterministic results are obtained. Using the KADABRA algorithm for betweenness centrality (a popular measure in network analysis) approximation as a case study, we demonstrate the empirical performance of our techniques. In particular, on a 32-core machine, our best algorithm is 2.9x faster than what we could achieve using a straightforward OpenMP-based parallelization and 65.3x faster than the existing implementation of KADABRA.

cs.DC↗

arXiv2018

Improving the betweenness centrality of a node by adding links

Betweenness is a well-known centrality measure that ranks the nodes according to their participation in the shortest paths of a network. In several scenarios, having a high betweenness can have a positive impact on the node itself. Hence, in this paper we consider the problem of determining how much a vertex can increase its centrality by creating a limited amount of new edges incident to it. In particular, we study the problem of maximizing the betweenness score of a given node -- Maximum Betweenness Improvement (MBI) -- and that of maximizing the ranking of a given node -- Maximum Ranking Improvement (MRI). We show that MBI cannot be approximated in polynomial-time within a factor $(1-\frac{1}{2e})$ and that MRI does not admit any polynomial-time constant factor approximation algorithm, both unless $P=NP$. We then propose a simple greedy approximation algorithm for MBI with an almost tight approximation ratio and we test its performance on several real-world networks. We experimentally show that our algorithm highly increases both the betweenness score and the ranking of a given node ant that it outperforms several competitive baselines. To speed up the computation of our greedy algorithm, we also propose a new dynamic algorithm for updating the betweenness of one node after an edge insertion, which might be of independent interest. Using the dynamic algorithm, we are now able to compute an approximation of MBI on networks with up to $10^5$ edges in most cases in a matter of seconds or a few minutes.

cs.DS↗

arXiv2018

Scalable Katz Ranking Computation in Large Static and Dynamic Graphs

Network analysis defines a number of centrality measures to identify the most central nodes in a network. Fast computation of those measures is a major challenge in algorithmic network analysis. Aside from closeness and betweenness, Katz centrality is one of the established centrality measures. In this paper, we consider the problem of computing rankings for Katz centrality. In particular, we propose upper and lower bounds on the Katz score of a given node. While previous approaches relied on numerical approximation or heuristics to compute Katz centrality rankings, we construct an algorithm that iteratively improves those upper and lower bounds until a correct Katz ranking is obtained. We extend our algorithm to dynamic graphs while maintaining its correctness guarantees. Experiments demonstrate that our static graph algorithm outperforms both numerical approaches and heuristics with speedups between 1.5x and 3.5x, depending on the desired quality guarantees. Our dynamic graph algorithm improves upon the static algorithm for update batches of less than 10000 edges. We provide efficient parallel CPU and GPU implementations of our algorithms that enable near real-time Katz centrality computation for graphs with hundreds of millions of nodes in fractions of seconds.

cs.DS↗

arXiv2018

Balanced k-means for Parallel Geometric Partitioning

Mesh partitioning is an indispensable tool for efficient parallel numerical simulations. Its goal is to minimize communication between the processes of a simulation while achieving load balance. Established graph-based partitioning tools yield a high solution quality; however, their scalability is limited. Geometric approaches usually scale better, but their solution quality may be unsatisfactory for `non-trivial' mesh topologies. In this paper, we present a scalable version of $k$-means that is adapted to yield balanced clusters. Balanced $k$-means constitutes the core of our new partitioning algorithm Geographer. Bootstrapping of initial centers is performed with space-filling curves, leading to fast convergence of the subsequent balanced k-means algorithm. Our experiments with up to 16384 MPI processes on numerous benchmark meshes show the following: (i) Geographer produces partitions with a lower communication volume than state-of-the-art geometric partitioners from the Zoltan package; (ii) Geographer scales well on large inputs; (iii) a Delaunay mesh with a few billion vertices and edges can be partitioned in a few seconds.

cs.DC↗