Fully Dynamic Algorithm for Top-k Densest Subgraphs

Size: px
Start display at page:

Download "Fully Dynamic Algorithm for Top-k Densest Subgraphs"

Transcription

1 Fully Dynamic Algorihm for Top-k Denses Subgraphs Muhammad Anis Uddin Nasir 1, Arisides Gionis 2, Gianmarco De Francisci Morales 3 Sarunas Girdzijauskas 4 Royal Insiue of Technology, Sweden Aalo Universiy, Finland Qaar Compuing Research Insiue, Qaar 1 anisu@kh.se, 2 arisides.gionis@aalo.fi, 3 gdfm@acm.org, 4 sarunasg@kh.se ABSTRACT Given a large graph, he denses-subgraph problem asks o find a subgraph wih maximum average degree. When considering he op-k version of his problem, a naïve soluion is o ieraively find he denses subgraph and remove i in each ieraion. However, such a soluion is impracical due o high processing cos. The problem is furher complicaed when dealing wih dynamic graphs, since adding or removing an edge requires re-running he algorihm. In his paper, we sudy he op-k denses-subgraph problem in he sliding-window model and propose an efficien fully-dynamic algorihm. The inpu of our algorihm consiss of an edge sream, and he goal is o find he node-disjoin subgraphs ha maximize he sum of heir densiies. In conras o exising sae-of-he-ar soluions ha require ieraing over he enire graph upon any updae, our algorihm profis from he observaion ha updaes only affec a limied region of he graph. Therefore, he op-k denses subgraphs are mainained by only applying local updaes. We provide a heoreical analysis of he proposed algorihm and show empirically ha he algorihm ofen generaes denser subgraphs han sae-of-he-ar compeiors. Experimens show an improvemen in efficiency of up o five orders of magniude compared o sae-of-he-ar soluions. 1 INTRODUCTION Finding a subgraph wih maximal densiy in a given graph is a fundamenal graph-mining problem, known as he denses-subgraph problem. Densiy is commonly defined as he raio beween number of edges and verices, while many oher definiions of densiy have been used in he lieraure [7, 28, 30, 31]. The denses-subgraph problem has many applicaions, for example, in communiy deeion [11, 14], even deecion [2], link-spam deecion [18], and disance query indexing [1]. In applicaions, we are ofen ineresed no only in one denses subgraph, bu in he op-k. The op-k denses subgraphs can be verex-disjoin, edge-disjoin, or overlapping [6, 17]. Differen objecive funcions and consrains give rise o differen problem formulaions [6, 17, 32]. In his work, we choose o maximize he sum of he densiies of he k subgraphs in he soluion. In addiion, Permission o make digial or hard copies of all or par of his work for personal or classroom use is graned wihou fee provided ha copies are no made or disribued for profi or commercial advanage and ha copies bear his noice and he full ciaion on he firs page. Copyrighs for componens of his work owned by ohers han ACM mus be honored. Absracing wih credi is permied. To copy oherwise, or republish, o pos on servers or o redisribue o liss, requires prior specific permission and/or a fee. Reques permissions from permissions@acm.org. CIKM 17, November 6 10, 17, Singapore. 17 ACM. ISBN /17/11...$15.00 DOI: hps://doi.org/ / (a) Figure 1: For he graph in Figure 1a, we are ineresed in exracing he op-3 denses subgraphs. Consider he arrival of an edge shown in red. Figure 1b shows he op-3 denses subgraphs afer he arrival. The objecive is o design an algorihm ha can efficienly mainain he denses subgraphs while keeping he number of updaes very low, in his case updaing only he verices in red. we seek a soluion wih disjoin subgraphs. This version of he problem is known o be NP-hard [6]. To complicae he maer, mos real-world graphs are dynamic and rapidly changing. For insance, Facebook users are coninuously creaing new connecions and removing old ones, hus changing he nework srucure. Twier users produce poss a a high rae, which makes old poss less relevan. Given he dynamic naure of many graphs, here we focus on a sliding-window model which gives more imporance o recen evens [4, 12, 13]. Finding he opk denses subgraphs in a sliding window is of ineres o several real-ime applicaions, e.g., communiy racking [33], even deecion [26], sory idenificaion [2], fraud deecion [8], and more. We assume he inpu o he sysem arrives as an edge sream, and seek o exrac he k verex-disjoin subgraphs ha maximize he sum of densiies [6]. A naïve soluion involves execuing a saic algorihm for he denses-subgraph problem k imes, while removing he denses subgraph in each ieraion. However, such a soluion is impracical as i requires o execue he algorihm k imes for each updae. An alernaive soluion o our problem is o use a dynamic denses subgraph algorihm in a pipeline manner, where he oupu of an algorihm insance serves as inpu o he following one. In his case, he graph and he insances of he algorihm are replicaed independenly across k insances of he algorihm, resuling in a high memory and processing cos. In his paper, we propose a fully-dynamic algorihm ha finds an approximae soluion. The proposed algorihm follows a greedy approach and updaes he densiies of he subgraphs conneced o verices affeced by edge operaions (addiion and removal). The (b)

2 algorihm is efficienly designed based on key properies of denses subgraphs, and i is compeiive agains oher recen dynamic algorihms [9, 15, 24]. Firs, our algorihm relies on he observaion ha only highdegree verices are relevan for he soluion. As many naural graphs have a heavy-ailed degree disribuion, he number of highdegree verices in a graph is relaively smaller han he number of low-degree ones. This simple observaion enables pruning a major porion of he inpu sream on-he-fly. Second, he verices ha are par of a denses subgraph are conneced srongly o each oher and weakly o oher pars of he graph. This enables independenly mainaining and locally updaing muliple subgraphs. Figure 1 provides an example which demonsraes his inuiion. The algorihm racks muliple subgraphs on-he-fly wih he help of a newly defined daa srucure called snowball. These subgraphs are sored in a bag, from which he k subgraphs wih maximum densiies are exraced. The algorihm runs in-place, and does no require muliple copies of he graph, hus making i memoryefficien. The one-pass naure of he algorihm allows exracing op-k denses subgraphs for larger values of k. We provide a heoreical analysis of he proposed algorihm, and show ha he algorihm guaranees 2-approximaion for he firs denses subgraph (k = 1) while providing a high-qualiy heurisic for k > 1 compared o oher soluions. Experimenal evaluaion shows ha our algorihm ofen generaes denser subgraphs compared o he sae-of-he-ar algorihms, due o he fac ha i mainains disconneced subgraphs separaely. In addiion, he algorihm provides improvemen in runime up o hree o five orders of magniude compared o he sae-of-he-ar. In summary, we make he following conribuions: We sudy he op-k denses verex-disjoin subgraphs problem in he sliding-window model. We provide a brief survey on adaping several algorihms for denses subgraph problem for he op-k case. We propose a scalable fully-dynamic algorihm for he problem, and provide a deailed analysis of i. The algorihm is open source and available online, ogeher wih he implemenaions of all he baselines. 1 We repor a comprehensive empirical evaluaion of he algorihm in which i significanly ouperforms previous sae-ofhe-ar soluions by several orders of magniude, while producing comparable or beer qualiy soluions. 2 PRELIMINARIES In his secion, we presen our noaion, revisi basic definiions, and formulae he op-k denses subgraphs problem. Consider an undireced graph G = (V, E) wih n = V verices and m = E edges. The neighborhood of v V is defined as N (v) = {u (v,u) E}, and is degree as d(v) = N (v). For a subse S V we define E(S) o be he se of edges whose boh endpoins are in S, and G(S) = (S, E(S)) he subgraph induced by S. The inernal degree of a verex v wih respec o S V is defined by d S (v) = N (v) S. 1 hps://gihub.com/anisnasir/topkdensessubgraph Finally, for a subse of verices S V we define is densiy ρ S by ρ S = E(S). (1) S Noe ha he densiy of any subgraph is equal o half of is average inernal degree. Definiion 2.1 (Denses subgraph). Given an undireced graph G = (V, E), he denses subgraph S is a se of verices ha maximizes he densiy funcion, i.e., S = arg max S V ρ S. (2) We say ha an algorihm A compues an α-approximaion of he denses subgraph if A compues a subse S V such ha ρ S 1 α ρ S, where S V is he denses subgraph of G. Nex we inroduce oher conceps relaed o denses subgraph: graph core, core decomposiion, and induced core subgraph of a verex. Definiion 2.2 (j-core). Given an undireced graph G = (V, E) and an ineger j, a j-core of G is a subse of verices C V so ha each verex v C has inernal degree d C (v) j, and C is maximal wih respec o his propery. Definiion 2.3 (Core decomposiion). A core decomposiion of a graph G = (V, E) is a nesed sequence { C i } of cores V = C 0 C 1... C l, (3) where each C i is a j-core for some j. Definiion 2.4 (Core number). Given a core decomposiion V = C 0 C 1... C l of a graph G = (V, E), he core number κ(v) of a verex v is he larges j such ha v C and C is a j-core. By overwriing noaion, he core number κ(c) of a core C is he larges j for which C is a j-core. Addiionally, we use κ S (v) o denoe he core number of a verex v in he subgraph induced by S. The larges core (or main core) of a subgraph of G(S) = (S, E(S)) is denoed by C l (S), while he main core of G is simply denoed by C l. Noe ha he densiy of a j-core is a leas j /2, as each verex in he core has degree a leas j and each edge is couned wice. This observaion implies ha he main core of a graph is a 2-approximaion of is denses subgraph. Lemma 2.5. Consider he core decomposiion of a graph G, i.e., C 1 C 2... C l. The maximum core C l is 2-approximaion of he denses subgraph of G [21]. Proof. Le S be he denses subgraph of G having densiy ρ S. Every verex in G(S ) has degree a leas ρ S ; oherwise a verex wih degree smaller han ρ S can be removed o obain an even denser subgraph. Thus, S is a ρ S -core. Given he core decomposiion of G, we know ha ρ Cl 1 2 κ(c l). We wan o show ha ρ Cl 1 2 ρ S. Assume oherwise, i.e., ρ C l < 1 2 ρ S. Then κ(c l ) < ρ S. I follows ha S is a higher-order core han C l, a conradicion. The concep of a core subgraph I(v) induced by a verex v [23, 27] is also perinen o our analysis.

3 Definiion 2.6 (Induced core subgraph). Given a graph G = (V, E) and a verex v V, he induced core subgraph of verex v, denoed by I(v), is a maximal conneced subgraph conaining he verex v s.. he core number of all he verices in I(v) is equal o κ(v). In oher words, he induced subgraph conains all verices ha are reachable from v and have he same core number κ(v). All previous definiions apply o saic graphs. Le us now focus on dynamic graphs. In paricular, we consider processing a graph in he sliding window edge-sream model [13]. According o his model, he inpu o our problem is a sream of edges. The edge e i is he i-h elemen of he sream. Equivalenly, we say ha edge e i has imesamp i. A sliding window W (x), defined a ime and of size x, is he se of all edges ha arrive beween e x+1 and e, W (x) = {e i, i [ x + 1, ]}. (4) For each edge e i = (u,v), we consider ha u and v appear a ime i, and we use V (x) o denoe he se of verices ha appear in a lengh-x sliding window a ime. The graph in a lengh-x sliding window a ime is hen defined o be G (x) = (V (x),w (x)). We are now ready o formally define he problem ha we consider in his paper, i.e., finding he op-k denses subgraphs in sliding window. We firs define he problem in a saic seing. Definiion 2.7. Given an undireced graph G = (V, E) and an ineger k > 0, he op-k denses subgraphs of G is a se of k disjoin maximal se of verices S = {S 1,..., S k } ha maximize he sum of is densiies: k ρ k (S) = max ρ Si, for all S i S subjec o i=1 here is no S j S i ρ Sj ρ Si, for all S i, S j S (5) S i S j =, for all i, j {1... k}, i j. (6) As already shown by Balalau e al. [6], he problem defined aboce is NP-hard, for any k > 1. The problem we consider in his paper is he following. Problem 2.8. Given a graph sream {e i } and a sliding window lengh x, mainain he op-k denses subgraphs ρ k (S) of he graphg (x), a any given ime. 3 BACKGROUND In his secion we presen a brief review over several algorihms for finding dense subgraphs. Addiionally, we discuss how hese mehods can be used for solving Problem 2.8. Denses subgraph in saic graphs. Finding he denses subgraph according o he densiy definiion (1) can be solved in polynomial ime. An elegan soluion involving reducion o he minimumcu problem was given by Goldberg []. As he fases algorihm o solve he minimum-cu problem runs in O(nm) ime [25], Goldberg s algorihm is no scalable o large graphs. Asahiro e al. [3] and Charikar [10] propose a linear-ime algorihm ha provides a facor-2 approximaion. This algorihm ieraively removes he verex wih he lowes degree in each ieraion, unil lef wih an empy graph. Among all subgraphs considered during his verex-removal process, he algorihm reurns he denses. The ime complexiy of his greedy algorihm is O(m + n). Bahmani e al. [5] propose a MapReduce version of he greedy algorihm, wih approximaion raio 2(1 +ϵ), while making O(log 1+ϵ n) passes over he inpu graph. Core decomposiion in saic graphs. The core decomposiion of a graph G is he process of idenifying all cores of G, as defined in 2.3. Baagelj and Zaversnik [7] propose a linear-ime algorihm o obain he core decomposiion. The algorihm firs considers he whole graph and hen repeaedly removes he verex wih he smalles degree. The core number κ(v) of a verex v is se equal o he degree of v a he momen ha v is removed from he graph. Denses subgraph in evolving graphs. There is a growing body of lieraure on finding dense subgraphs in evolving graphs [9, 15, 19, 24]. We focus mainly on he deerminisic algorihm for denses subgraph in evolving graphs. For insance, Epaso e al. [15] propose an efficien algorihm for compuing he denses subgraph in he dynamic graph model [16]. Their work assumes ha edges are insered ino he graph adversarially bu deleed randomly. Even hough he algorihm can, in pracice, handle arbirary edge deleions, is approximaion guaranees hold only under he random edge-deleion assumpion. The algorihm is similar o he one by Bahmani e al. [5], and i provides a 2(1 + ϵ) 6 -approximaion of he denses subgraph, while requiring polylogarihmic amorized cos per updae wih high probabiliy. Core mainenance in evolving graphs. Saríyüce e al. [27] propose he raversal algorihm, for efficien core mainenance. This algorihm idenifies a small se of verices ha are affeced by edge updaes and processes hese verices in linear ime in order o mainain a valid core decomposiion. Li e al. [23] propose an efficien hree-sage algorihm for core mainenance in large dynamic graphs. The algorihm mainains a core decomposiion of an evolving graph by applying updaes o very few verices in he graph. Once hese few verices have been idenified, he algorihm compues he correc core numbers via a quadraic operaion. Finding op-k denses subgraphs. The problem of finding op-k denses subgraphs has been mainly sudied for finding overlapping subgraphs in saic graphs [6, 17]. Nex, we discuss how he algorihms presened above (Charikar [10], Baagelj and Zaversnik [7], Bahmani e al. [5], Saríyüce e al. [27], Li e al. [23], and Epaso e al. [15]) can be used o produce op-k denses subgraphs. Our firs observaion is ha a se of k dense subgraphs can be obained from any algorihm ha finds he denses subgraph by k repeaed invocaions. The ime complexiy of compuing a se of k dense subgraphs in his manner is simply he runningime complexiy of he denses-subgraph algorihm muliplied by k. From a pracical poin of view, all he saic algorihms menioned are no in-place algorihms, and hus require copying he whole graph for processing. Furhermore, when a verex or edge is added or deleed from he graph, he whole k dense subgraph compuaion has o be repeaed. The second observaion is ha, by using he algorihm of Epaso e al. [15], we can obain a se of k dense subgraphs by running k insances of he fully-dynamic algorihm in a pipeline manner. The idea is o run k insances of he algorihm in which he oupu of each insance i {0... k 1} is fed ino he nex (i + 1) insance as a removal operaion. The pipeline version of he algorihm requires keeping k copies of he inpu graph and an addiional

4 O(kn) size space for bookkeeping. Noe ha he oupu of each insance of he pipeline migh cascade, which requires updaing he verices in all he insances. In paricular, verices ha cease o be par of soluion in upsream insances need o be added in downsream insances. Likewise he verices ha become par of denses subgraphs in upsream insances need o be removed from a downsream insances. The modificaion of he algorihm, as discussed above, is expensive in erms of memory, as i requires replicaing he graph and he algorihm s srucures k imes. In addiion, running and mainaining k parallel insances makes he algorihm compue-inensive. Finally, o mainain op-k denses subgraphs in evolving graphs, we can leverage algorihms for core decomposiion mainenance [23, 27]. by leveraging Lemma 2.5, Thus, he idea is o find and mainain op-k disjoin maximum cores. In order o mainain such cores we run a single insance of he algorihm by Saríyüce e al. [27] or Li e al. [23] ha mainains he core number of all he verices in he graph. We hen exrac he op-k denses subgraphs by: (i) exracing he main core, (ii) removing he verices in he main core and updaing he core number for res of he verices, and (iii) repeaing he seps unil k subgraphs are exraced. 4 ALGORITHM The main idea of our algorihm is o mainain and updae muliple dense subgraphs online. These subgraphs are candidaes for he op-k denses subgraphs. However, mainaining muliple subgraphs for fully-dynamic sreams requires answering wo ineresing quesions: (i) how o reduce he search space of he soluion, and (ii) how o spli he whole graph ino subgraphs. To answer he aforemenioned quesions, we make wo observaions. Firs, since dense subgraphs are formed by relaively highdegree verices, one can find dense subgraphs by keeping rack of hese high-degree verices only. Second, hese subgraphs can be updaed locally upon edge updaes, wihou affecing he oher pars of he graph. Based on hese observaions, we develop an algorihm ha reduces he soluion space by considering only high-degree verices, and divides he whole graph ino smaller subgraphs, each represening a dense subgraph. The op-k denses subgraphs among he candidae subgraphs provide a soluion for Problem 2.8. We begin by designing an algorihm o find he denses subgraph (op-1) and hen we exend i o find he op-k denses subgraphs. Our algorihm migh no be he mos efficien soluion for he (op-1) denses-subgraph problem per se, bu i provides efficien oucomes when exended o solve he op-k denses-subgraph problem. We sar by defining some properies of he denses subgraph ha we leverage in our algorihm. Lemma 4.1. Given an undireced graph G = (V, E), he denses subgraph S V wih densiy ρ S, all he verices v S have degree d S (v) ρ S. Proof. This lemma holds according o he definiion of opimal densiy. In an opimal soluion, each verex has degree larger han or equal o ρ S. Oherwise, removing he verex from he subgraph will increase he average degree, and hus he densiy, of he subgraph. Given Lemma 4.1, a any ime, he denses subgraph S of graph G conains only verices v ha have degree d(v) d S (v) ρ S. Then, given G and ρ S, we wan o compue he denses subgraph afer he addiion of a new verex u V a ime + 1. Le d(u) be he degree of verex u V +1 and S+1 be he denses subgraph a ime + 1. For simpliciy, assume ha he graph G +1 is conneced. According o Lemma 4.1, for any verex u o be he par of he denses subgraph, is inernal degree saisfies d S +1 (u) ρ S +1. As verex u is added o he graph he new densiy is always greaer, i.e., ρ S +1 ρ S. Therefore, for verex u o be he par of denses subgraph, he degree of verex u should saisfy d(u) ρ S. Therefore, if he degree of verex u is lower han he ρ S, i canno be par of he denses subgraph ρ S +1 and can be ignored. Now, considering he case when d(u) ρ S. Adding he verex u o denses subgraph will updae he densiy: ρ S +1 E(S ) + d S (u) S +1. (7) We also know ha ρ S +1 ρ S, which means ha E(S ) + d S (u) S +1 E(S ) S. (8) From his inequaliy i follows d S (u) ρ S. Using hese properies, we ignore he verices of he new edge ha have degree lower han he curren esimae of he densiy. Furher, we are ineresed in finding he main core in he remaining subgraph of high-degree verices, as i represens a 2- approximaion of he denses subgraph according o Lemma 2.5. To his end, we propose a new daa srucure ha relies on Lemma 4.1, he snowball. 4.1 Snowball A snowball D is an incremenal daa srucure ha sores a srongly conneced subgraph, which mainains he following invarians: The core number κ D (v) of each verex v D inside a snowball is equal o he main core (C l (D)) of he snowball. All he verices in he snowball are conneced. These invarians ensure ha all he verices in he snowball have he same core number, which is he main core of he snowball by definiion. A snowball mainains hese invarians while handling he following graph updae operaions: a) adding/removing a verex, and b) adding/removing an edge. 4.2 Bag of snowballs The high-degree verices in he graph are assigned o a snowball. As hese verices are no srongly conneced, hey migh end up in differen snowballs. We sore each of hese disconneced snowballs in a daa srucure called he bag, denoed by B. The bag ensures ha each snowball is verex disjoin. Furher, he bag provides an addiional operaion: exracing he denses snowball among he se of snowballs. The densiy of he exraced snowball is he maximal densiy, which is he hreshold separaing he high-degree verices from he low-degree ones. We denoe his esimae of he maximal densiy ρ S. The bag is a supergraph which conains a se of snowballs and all he edges beween he snowballs. We mainain all he core numbers of he nodes in he bag by leveraging a core decomposiion

5 Bag Bag D 1 D 2 D 1 D 2 Figure 2: Example showing ha he bag requires mainaining he core number of he verices. Iniially, he bag conains wo snowballs wih core number 2, i.e., D 1 and D 2. Consider he arrival of he edges shown in red. The greedy assignmen of he edges migh skip creaing a new snowball wih core number 3, using he four nodes in he middle. algorihm (see Secion 3). The core number of each node in he bag is used o ensure ha each node has he maximum possible core number. Figure 2 provides an example explaining one of he issues ha may arise. In he example, he bag conains wo snowballs, however, i is possible o produce a new snowball wih a larger core number. Nex, we define he algorihms o updae his daa srucure upon graph updaes. 4.3 Addiion operaions Verex addiion: As discussed in Secion 2, he updaes appear in he form of an edge sream. Here, we define he verex addiion algorihm ha acs as a helper for edge addiion. The algorihm is riggered when a leas one of he endpoins of a new edge is a high-degree verex. In paricular, here are wo cases o consider: 1) he bag already conains he high-degree verex, and 2) he bag does no conain he high-degree verex. In boh cases, he goal is o add he new verex o one of he snowballs (if needed). Algorihm 1 defines he algorihm for verex addiion. For he firs case, he algorihm scans he bag o find he snowball ha conains he verex and reurns i. For he second case, he algorihm firs idenifies he candidae snowballs, hen i assigns he verex o one of he candidae snowballs. The candidae snowballs are he ones having he main core number smaller han or equal o he inernal degree of he new verex (κ u (D) C l (D)). Among he candidae snowballs, he new node is assigned o he snowball wih maximum inernal degree d u (D), breaking ies randomly. Once a verex is added o a snowball, he core number of he snowball may increase. This change requires removing he verices wih core number lower han he main core of he snowball. This procedure can be implemened efficienly in linear ime by soring he verices based on heir degree similar o bin sor. 2 Verificaion. Due o he greedy assignmen of verex o he snowball, i is possible ha he verex ends up no having he highes possible core number. For example, Figure 3 shows an example where he greedy assignmen does no resul in opimal soluion. Therefore, afer addiion, he algorihm ensures ha he core number of he snowball, where he new verex is added, equals he core number of he new verex in he graph. The algorihm verifies ha he core number of he added verex by comparing i wih he core number of he verex in he bag. Noe ha he bag represens he supergraph conaining all he snowballs and edges beween he snowballs. If he core number wihin he bag is larger han he one in he snowball, he algorihm merges all he 2 The MainainInvarian mehod a line 12 of Algorihm 1. Figure 3: Example showing ha arrival of a new edge allows he verex ha is par of snowball D 2 o become par of snowball D 1, which has greaer core number (3). Algorihm 1 Verex Addiion in he Bag of Snowballs 1: procedure addobag(u) 2: S 3: for D i B do 4: if u D i hen 5: reurn D i Firs Case 6: if (d Di (u) C l (D i ) and ρ S < ρ Di ) hen 7: S D i 8: if S = hen 9: S {u } 10: else 11: S S {u } 12: MainainInvarian(S ) 13: reurn S Second Case Algorihm 2 Mainain Invarian 1: procedure MainainInvarian(D i ) 2: do 3: repea f alse 4: for u D i do 5: if ((κ Di (u) < C l (D i )) hen 6: D i D i \{u } 7: if (d(u) ρ S ) hen 8: addobag(u) 9: repea rue 10: while repea snowballs in he induced subgraph of newly added verex. As all he verices in he induced subgraph have he same core number, merging hem ensures creaing a larger snowball. 3 We leverage he core decomposiion algorihm by Saríyüce e al. [27] for he implemenaion. Algorihm 3 Fix Main Core 1: procedure fixmaincore 2: for D i B do 3: if (D i I(u) > 0) hen 4: D u D u D i 5: MainainInvarian(D u ) Theorem 4.2. Given he bag B, he algorihm ensures ha B conains he main core of he graph wihin one of he snowballs afer he verex addiion. Proof. Le us assume ha a ime he bag conains he main core of he graph. Now, we need o show ha a ime + 1, afer he node addiion, he bag sill conains he main core of he graph. In general, verex addiion mehod is called whenever here is an edge addiion. The only way for he new verex o affec he main core of he graph is ha he new verex is he par of he main core. 3 The FixMainCore mehod a line 15 of Algorihm 4.

6 Algorihm 4 Edge Addiion 1: procedure addedge((u,v)) 2: if ((d(u) < ρ S )) and (d(v) < ρ S )) hen 3: reurn 4: else if ((d(u) ρ S ) and (d(v) < ρ S )) hen 5: D u addobag(u) 6: else if ((d(u) < ρ S ) and (d(v) ρ S ) hen 7: D v addobag(v) 8: else 9: D u addobag(u) 10: D v addobag(v) 11: if (D u = D v ) hen 12: D u D u (u, v) 13: MainainInvarian(D u ) 14: else if (v I(u)) hen 15: fixmaincore(u) Afer he addiion of he verex in he bag, he algorihm verifies he core number by comparing he core number of he verex in he bag and he snowball. If he core number of he verex in he bag is greaer, he algorihm merges he snowballs conaining he verices in he induced graph of he new node in he bag. This creaes a new snowball wih a greaer core number. Edge addiion: In his case, he sae of he bag is only affeced if a leas one of he verices in he new edge is a high-degree verex. In paricular, here are wo cases o consider: a) only one of he verices is a high-degree verex and b) boh he verices are highdegree verices. For he firs case, he algorihm leverages he verex addiion mehod o add he verex o he bag of snowballs. For he second case, when boh verices are added o he bag of snowballs, he algorihm verifies ha he main core exiss in he bag. When boh verices are added o he same snowball, he algorihm adds he new edge o he same snowball and ensures ha he invarian holds. Conversely, when he wo verices are added o wo differen snowballs, he algorihm verifies if he verices exis in each ohers induced subgraphs and fixes he main core for boh he verices. Algorihm 4 describes he algorihm for edge addiion. Theorem 4.3. Algorihm 4 mainains he main core of he graph in one of he snowballs inside he bag. Proof. The proof for all he cases, oher han he case when boh he verices of he new edge are assigned o wo differen snowballs, is similar o he verex addiion algorihm. Therefore, we consider he case when boh end verices of he added edge are added o wo differen snowballs. For his case, we rely on he graph in he bag. We check if boh verices are in he same core graph in he bag, and fix he core number of he verices if hey belong o he same induced subgraph. This ensures creaing he graph wih he highes core number. 4.4 Removal operaions Verex removal: Similarly o he addiion operaions, we firs define he procedure for removing a verex from he bag of snowballs. The verex removal mehod is used as a subrouine for he edge removal process. A verex is only removed from he bag when is degree becomes lower han he maximal densiy. Therefore, according o Lemma 2.5, he removed verex canno be par of he main core. The algorihm removes he verex from he snowball wihin a bag wihou doing any oher operaion. Algorihm 5 Edge Deleion 1: procedure removeedge((u,v)) 2: if ((d(u) < ρ S ) and (d(v) < ρ S )) hen 3: reurn 4: if ((d(u) < ρ S ) and (d(u) + 1 ρ S )) hen 5: removeverex(u) 6: reurn 7: if ((d(v) < ρ S ) and (d(v) + 1 ρ S )) hen 8: removeverex(v) 9: reurn 10: for D i B do 11: if (D i (u, v) ) hen 12: D i D i \(u, v) 13: for x D i do 14: if (κ B (x) > κ Di (x)) hen 15: fixmaincore(x) 16: MainainInvarian(D i ) Edge removal: Now we urn our aenion o edge deleion, which follows he same paern as edge addiion. Again, we leverage he bag o ensure ha here exis a snowball wih a core number equal o he main core of he graph. Algorihm 5 shows he algorihm for edge deleion. The bag does no require any updae when eiher one of he verices is low-degree or if boh he verices belong o wo differen snowballs. Therefore, we consider he case when one of he verices lie a he boundary of high-degree verices. Tha is, he edge deleion moves he verex from he high-degree o low-degree. In his case, he algorihm only requires removing he verex from he bag, wihou performing any oher operaions. The ineresing case is where boh verices of he deleed edge are high-degree and belong o he same snowball. In his case, he algorihm removes he edge from he snowball. Furher, i verifies and updaes (if needed) he core number of he verices affeced by he updae in he snowball. Lasly, edge deleion migh reduce he maximal densiy, and hus require adding o he bag new verices whose degree is now greaer han he new maximal densiy. Theorem 4.4. Given he bag conaining a snowball wih he same core number as he main core of he bag, afer he edge deleion, Algorihm 5 mainains he main core of he graph in one of he snowballs inside he bag. Proof. An edge removal affecs he bag of snowballs only when boh he verices corresponding o he removed edge belong o he same snowball. In his case, edge removal migh reduce he core number of all he verices in he snowball. The algorihm ensures ha he verices in he snowball have heir maximum possible core number by comparing heir core number in he snowball wih he core number in he bag. Therefore, by verifying and fixing he core numbers, he algorihm mainains he main core of he graph in one of he snowballs inside he bag. 4.5 Fully-dynamic op-k denses subgraphs Now ha we have a fully dynamic algorihm for finding he denses subgraph in sliding windows, we move our aenion o he op-k denses-subgraph problem. Le ρ S ρ 1... ρ z 1 represen he densiies of he z subgraphs in he bag, where ρ S is he densiy of he denses subgraph. The bag conains he verices ha have a degree greaer han he densiy ρ S. To ensure ha he bag conains a leas k subgraphs,

7 (a) Figure 4: Example showing op-k version of he algorihm. The densiies of op-3 subgraphs, before he arrival of he new edge in he bag are 1.8, 1.5 and The algorihm sores all he verices in he bag wih degree greaer han equal o 1.25, as shown in Figure 4b. Afer he arrival of he new edge, he updae only affecs one of he subgraphs in he bag. we modify he algorihm o keep all he verices wih degree greaer han ρ k 1. The only modificaion required is o replace ρ S by ρ k 1 in Algorihm 2, Algorihm 4 and Algorihm 5. Furher, we leverage he prioriy queue o exrac ρ k 1 from he bag of snowballs. This simple modificaion ensures ha he bag conains a leas k subgraphs and enables accessing he op-k denses subgraphs in he sliding window model. Noe ha he new definiion of high-degree verices is relaed o he densiy of he k-h op denses subgraph. The modified algorihm guaranees ha he bag conains he verices wih a degree greaer han he ρ k 1 afer any graph updae. These graph updaes include boh edge addiions and removals. I is necessary for he algorihm o consider edge addiions, as hey migh affec he value of ρ k 1 by merging muliple subgraphs. Similarly, edge deleions have o be considered, as hey migh affec he value of ρ k 1 by reducing he densiy of any of he op-k denses subgraphs, merging muliple subgraphs, or spliing a subgraph. As he algorihm ensures keeping he main core in he bag, i guaranees 2-approximaion for he firs denses subgraph (k = 1) while providing a high-qualiy heurisic for k > 1 compared o oher soluions. We provide an example in Figure 4 for he op-k denses subgraph algorihm by expanding on he Figure 1a. We conclude his secion wih he heorem ha provides a bound for our proposed algorihm. These bounds can be generalized for any algorihm ha adaps o a soluion for denses subgraph problem for he op-k case (see secion 3 for examples). Theorem 4.5. Given an ineger k, he bag conains a se of subgraphs ha provides 2k-approximaion of he op-k denses-subgraph problem. Proof. We know ha he graph does no conain any subgraph wih densiy greaer han he opimal densiy (ρ S ). Therefore, we know ha he sum of densiies for he op-k denses-subgraph problem is upper bounded by k ρ S. Furher, he bag conains ρ S, which provides a 2-approximaion of he denses subgraph. This implies ha he sum of densiies of op-k denses subgraph in he bag 1 2 ρ S. Puing above wo observaions ogeher, we can clearly see ha he bag provides a soluion ha is a 2k-approximaion for he problem. (b) 4.6 Daa srucure The proposed algorihm requires accessing he neighborhood informaion of every node. Specifically, we are ineresed in performing hree queries on a given verex: a) exrac is degree, b) exrac all is neighbors, and c) given densiy ρ S, exrac all he verices wih degrees greaer han he densiy ρ S. Verex map: To answer he firs wo queries, we need o sore he neighborhood informaion for all verices. We sore he informaion in a hashmap wih keys being he verex idenifiers and values being he neighbors of each verex. Verex map allows performing search and updae operaions in amorized consan ime. Degree able: For he hird query, we need o order he verices by heir degrees. We use bin sor o order he verices by heir degrees, which enables exracing he verices wih degrees greaer han he densiy ρ S in consan ime. 5 DISCUSSION Mos of he soluions ha leverage a saic algorihms [5, 7, 10] require ieraing over he enire graph k imes upon any updae. Comparaively, our algorihm mosly ouches a limied region in he graph for any updaes, which makes our algorihm perform significanly faser. The op-k denses subgraph algorihm ha adaps o Epaso e al. [15] requires replicaing he graph across he k insances. In addiion, updaes across k insances migh cascade and require running he algorihm k imes, similarly o he saic op-k soluions. Our algorihm ouperforms he pipeline version of he op-k algorihm boh in erms of memory and compuaion. Finally, incremenal algorihms for k-core mainenance requires removing op-k denses subgraph upon every updae in he graph. Each removal of a denses subgraph requires updaing he core number of he remaining verices. Our algorihm efficienly avoids his removal sep by mainaining he subgraphs during execuion. Performance. The proposed algorihm leverages he skew in realworld graphs and ignores a major porion of he inpu sream on he fly. In addiion, high-degree verices are sored as small subgraphs so ha each updae is ofen applied locally on hese subgraphs. This design allows our algorihm o operae on jus small porions of he graph for each updae, raher han ieraing on he whole graph. Finally, we use he k-core algorihm [23, 27], which allows each high-degree node o updae heir core number wih a complexiy independen of he graph size. These improvemens enable our algorihm o perform significanly beer han he oher sae-ofhe-ar sreaming algorihms. Memory. Our algorihm requires only O(n 2 polylogn) memory, compared o O(kn 2 polylogn) for he op-k version of [15] and O(n 2 polylogn) for he oher algorihms discussed in Secion 3. Tigh Bounds. We show via an example ha any algorihm for denses subgraph problem can only produce a k-approximaion for op-k denses subgraph problem. Consider a graph G ha conains n m verices, for any n k. The n nodes connec o form a circle. In he circle, here is one edge connecing wo non-adjacen verices. Addiionally, each of he n nodes connecs o exacly m neighbors. The inner circle of he graph G is he denses subgraph. Removing he denses subgraph leaves he res of he verices compleely disconneced. Hence, he sum of densiy will be equal o one. Now,

8 Table 1: Daases used in he experimens. Daase Symbol n m d(v) Amazon [22] AM DBLP 1 [22] DB Youube [22] YT DBLP 2 4 DB Live Journal 5 LJ Orku [22] OT Friendser [22] FR consider he case when he denses subgraph is no removed firs. In his case, each node in he circle along wih is m neighbors creae a subgraph of densiy almos equal o one, for higher values of m. Therefore, one can reurn k such subgraphs wih he sum of densiies equal o k, which is k imes beer han he previous case. 6 EVALUATION We conduc an exensive empirical evaluaion of he proposed algorihm, and provide comparisons wih he exising soluions. In paricular, we answer he following quesions: Q1: Wha is he impac on he qualiy of subgraphs? Q2: Wha are he gains in performance? Q3: How does he algorihm perform in erms of differen inpu parameers? 6.1 Experimenal seup Daases. Table 1 shows he daases used in he experimens. The daases are seleced due o heir public availabiliy. We evaluae all he algorihms in he sliding window model. The number of edges in he sliding window is an inpu parameer. Merics. We evaluae he qualiy and efficiency of he algorihms. We assess he qualiy by he objecive funcion, i.e., he sum of densiies of he subgraphs produced by an algorihm. We evaluae he efficiency of an algorihm by reporing he average updae ime and he memory usage. The average updae ime is he average ime i akes o move he sliding window. This includes adding he new edge, removing he oldes edge, and updaing he op-k denses subgraphs. We repor he memory usage as he average percenage of occupied memory. Algorihms. Table 2 shows he noaions used for differen algorihms. The algorihms by Bahmani e al. [5] and Epaso e al. [15] require an addiional epsilon parameer for execuion, which provides a rade-off beween qualiy and execuion ime. Here we use he defauls proposed by he auhors. Sream ordering. We consider wo commonly used sream ordering schemes [29]: BFS: The ordering is a resul of a breadh-firs search saring from a random verex. DFS: The ordering is a resul of a deph-firs search saring from a random verex. Experimenal environmen. We conduc our experimens on a machine wih 2 Inel Xeon Processors E and 128GiB of 4 hp://konec.uni-koblenz.de/neworks/dblp coauhor 5 hp://konec.uni-koblenz.de/neworks/livejournal-links Table 2: Noaion for he op-k algorihms. Symbol Reference Algorihm Top-1 Approx. In-place Guaranees CH Charikar [10] 2 No V B Baagelj and Zaversnik [7] 2 No BB ϵ Bahmani e al. [5] 2(1 + ϵ) No RL Li e al. [23] 2 Yes T R Saríyüce e al. [27] 2 Yes AE ϵ Epaso e al. [15] 2(1 + ϵ) 6 No GR his paper 2 Yes memory. All he algorihms are implemened in Java and execued on JRE 7 running on Linux. The source code is available online Experimenal resuls Q1: In his experimen, we compare he qualiy of he resuls produced by he differen algorihm as measured by he objecive funcion. In order o be able o run mos of he algorihms, we use he smaller daases, i.e., AM, DB 1, YT, and DB 2. As RL and TR produce same resuls in erms of qualiy, we only repor he resuls for one of hem. Due o space consrains, we show he resuls only wih he DFS ordering. However, we achieve similar resuls also wih he BFS one. We se he size of he sliding window x = 100k. For he saic algorihms, we execue hem in micro-baches in which he op-k denses subgraph is recompued afer 1k edge updaes, which gives hem a subsanial advanage. For BB ϵ and AE ϵ, we use ϵ = We se he maximum execuion ime o 7 days, due o which he AE ϵ is no able o finish on DB 2 and YT. Finally, RL and TR conain an updae phase due o he ieraive exracion of he op-k denses subgraph. This phase is expensive, as he main core consiss of high-degree verices. To alleviae his issue, we execue he exracion phase every 10k updae operaions. Figure 5 shows he qualiy resuls. Our algorihm GR achieves compeiive qualiy, ofen generaing denser subgraphs compared o all he oher algorihms. For example, for he AM daase, GR produces subgraphs ha are 1.5 imes beer han he bes algorihm among he sae-of-he-ar soluions. All he oher algorihms produce consisen resuls across all he daases. For insance, V B, RL, and TR produce op-k dense subgraphs of lowes qualiy compared o he oher algorihms. This resul is caused by he removal of he main core (backbone) of he graph in each of he k ieraions. In addiion, i shows how he problem considered in his paper, while relaed, is differen from a simple k-core decomposiion. The resuls also validae ha BB ϵ and AE ϵ provide weaker guaranees on he qualiy compared o CH. Q2: In his experimen, we urn our aenion o evaluae he efficiency of he proposed algorihm, as measured by he average updae ime and memory usage. We again selec several daases, i.e., AM, DB 1, YT, DB 2, and LJ, and execue CH, V B, BB, RL,TR, AE, and GR. We exclude he larges daases as only a few algorihms are able o handle hem. The algorihms are execued wih boh BFS and DFS ordering of he edge sream. The sliding window size is se o x = 100k edges, and k = 10 unless oherwise specified. 6 hps://gihub.com/anisnasir/topkdensessubgraph

9 Sum of densiies DB 1 CH VB GR BB 0.1 AE 0.1 TR Edge sream (DFS) AM Edge sream (DFS) DB Edge sream (DFS) YT Edge sream (DFS) Figure 5: Sum of densiies for op-10 denses subgraphs on he DB 1, AM, DB 2 and YT daases, sliding window size 100k. Table 3: Memory consumpion as a percenage of oal memory for algorihms wih DB 2 and LJ daase, sliding window size 100k. CH V B BB ϵ RL T R AE ϵ GR DB LJ Average updae ime (s) 10 3 DFS DB 1 AM DB 2 YT LJ VB CH BB 0.1 AL TR EP 0.1 GR Algorihm 10 3 BFS VB CH BB 0.1 AL TR EP 0.1 GR Algorihm Figure 6 shows he efficiency resuls (noe he logarihmic scale). The proposed algorihm, GR, ouperforms all oher ones in erms of updae ime for all he daases and boh ordering schemes. RL and T R are he slowes algorihms. In paricular, for DFS, our algorihm achieves a performance gain of five orders of magniude. This resul is noeworhy as even hough our algorihm is dependen on core decomposiion, i is sill able o bea a naïve applicaion of hose algorihms by a wide margin. This difference is due o he fac ha boh algorihms require mainaining he core number for all he verices. Addiionally, he exracion of op-k denses subgraph furher hampers heir efficiency. AlgorihmsCH, V B, and BB 0.01 perform unfavorably due o heir saic naure. In his case, our algorihm is able o achieve more han hree orders of magniude improvemen in efficiency. Algorihm AE 0.01 is bes performing among he baselines, and even ouperforms GR in one daase for one specific ordering (AM wih BFS). However, GR sill ouperforms AE 0.01 by nearly hree orders of magniude in mos oher cases. Finally, we observe he overall memory consumpion of all he algorihms (repored in Table 3). The memory requiremen of our algorihm lies beween he saic and he dynamic algorihms, i.e., AE 0.01 requires he larges amoun of memory, while he saic algorihms require he leas. These resuls are in line wih our expecaions from he discussion in Secion 3. Q3: In his experimen, we evaluae he scalabiliy of our algorihm. Firs, we execue GR wih differen values of k. Alongside, we repor he average updae ime for BB ϵ algorihm. We choose BB ϵ, raher han AE ϵ, as i can execue in mirco-baches, i.e., op-k denses subgraph every 100 edges. We se he sliding window size x = 1M, and use he YT and DB 2 daase wih DFS ordering. Figure 7 repors he average updae ime for boh algorihms. The average updae ime of GR remains consisen even for higher values of k, whereas he execuion ime of BB 0.01 increases wih he parameer. Second, we execue GR wih differen sizes of sliding window, i.e., x = 10k, 100k, 1M, and 10M. We se k = 10 for his experimen. Again, we use he YT and DB 2 daases wih DFS ordering. Figure 6: Updae ime for he algorihms on he DB 1, AM, DB 2, YT, and LJ daases, when using boh DFS and BFS ordering and k = 10. Average Updae Time (s) GR BB 0.1 YT k 10-2 Figure 7: Updae ime for GR and BB 0.01 on he YT and DB 2 daases as a funcion of k. Noe ha BB 0.01 is execued in micro-baches of size 100. Table 4: Average updae ime for he GR algorihm wih sliding windows of differen sizes x. x = 10k x = 100k x = 1M x = 10M YT 0.80ms 91.14ms 90.33ms 95.49ms DB ms 7.58ms 8.45ms 32.21ms Table 4 repors he average updae ime for differen configuraions. Increasing he size of he sliding window does no affec he average updae ime significanly. This resul validaes our claim ha mos of he updaes are local o he subgraphs, and do no require ieraing hrough he whole graph o exrac he op-k denses subgraphs. Finally, we sudy he performance of our algorihm in on he larges daases. We selec he OT and FR daases wih DFS ordering, and execue he algorihm wih a sliding window of x = 100k, wih k = 10. Figure 8 shows he resul of he experimen. The plo is generaed by aking he moving average of he updae ime and he sum of densiies. The average updae ime of he algorihm k DB 2

10 Average updae ime (s) average updae ime sum of densiies OT Edge sream (DFS) 10-4 average updae ime sum of densiies FR Edge sream (DFS) Figure 8: Qualiy and efficiency for GR on he OT and FR daases over he sream. mosly remains consan hroughou he execuion, and our algorihm provides seady efficiency over he sream. This behavior remains consisen even when he densiies are flucuaing, as in he case for he OT daase. 7 RELATED WORK Valari e al. [32] were he firs one o sudy he op-k denses subgraph problem for a sream consising of a dynamic collecion of graphs. They proposed boh an exac and an approximaion algorihm for op-k denses subgraph discovery. Similar o our algorihm, he proposed algorihm relies on he core decomposiion o provide he approximaion guaranees. The op-k denses subgraphs produced by he algorihm are edge-disjoin. Balalau e al. [6] sudied he problem of finding he op-k denses subgraph wih limied overlap. They defined he op-k denses subgraphs as a se of k subgraphs ha maximizes he sum of densiies, while saisfying an upper bound on he pairwise Jaccard coefficien beween he se of verices of he subgraphs. The problem of finding he op-k denses subgraph as sum of densiies was shown o be NP-hard [6] and efficien heurisic was proposed o solve he problem. Furher, Galbrun e al. [17] sudied he problem of finding he op-k overlapping denses subgraphs and provided consan-facor approximaion guaranees. 8 CONCLUSION We sudied he op-k denses subgraphs problem for graph sreams, and proposed an efficien one-pass fully-dynamic algorihm. In conras o he exising sae-of-he-ar soluions ha require ieraing over he enire graph upon updae, our algorihm mainains he soluion in one-pass. Addiionally, he memory requiremen of he algorihm is independen of k. The algorihm is designed by leveraging he observaion ha graph updaes only affec a limied region. Therefore, he op-k denses subgraphs are mainained by simply applying local updaes o small subgraphs, raher han he complee graph. We provided a heoreical analysis of he proposed algorihm and showed empirically ha he algorihm ofen generaes denser subgraphs han he sae-of-he-ar soluions. Furher, we observed an improvemen in performance of up o five orders of magniude when compared o he baselines. This work gives rise o furher ineresing research quesions: Is i necessary o leverage k-core decomposiion algorihm as a backbone? Is i possible o achieve sronger bounds on he hreshold for high-degree verices? Can we design an algorihm wih a space bound on he size of he bag? Is i possible o achieve sronger approximaion guaranees for he problem? We believe ha solving hese quesions will furher enhance he proposed algorihm, making i a useful ool for numerous pracical applicaions. Sum of densiies REFERENCES [1] Takuya Akiba, Yoichi Iwaa, and Yuichi Yoshida. 13. Fas exac shores-pah disance queries on large neworks by pruned landmark labeling. In SIGMOD. ACM, [2] Alber Angel, Nikos Sarkas, Nick Koudas, and Divesh Srivasava. 12. Dense subgraph mainenance under sreaming edge weigh updaes for real-ime sory idenificaion. VLDB 5, 6 (12), [3] Yuichi Asahiro, Kazuo Iwama, Hisao Tamaki, and Takeshi Tokuyama Greedily finding a dense subgraph. In SWAT [4] Brian Babcock, Shivnah Babu, Mayur Daar, Rajeev Mowani, and Jennifer Widom. 02. Models and issues in daa sream sysems. In PODS. ACM, [5] Bahman Bahmani, Ravi Kumar, and Sergei Vassilviskii. 12. Denses subgraph in sreaming and mapreduce. VLDB 5, 5 (12), [6] Oana Denisa Balalau, Francesco Bonchi, TH Chan, Francesco Gullo, and Mauro Sozio. 15. Finding subgraphs wih maximum oal densiy and limied overlap. In WSDM. ACM, [7] Vladimir Baagelj and Majaz Zaversnik. 03. An O (m) algorihm for cores decomposiion of neworks. arxiv preprin cs/ (03). [8] Alex Beuel, Wanhong Xu, Venkaesan Guruswami, Chrisopher Palow, and Chrisos Falousos. 13. Copycach: sopping group aacks by spoing locksep behavior in social neworks. In WWW [9] Sayan Bhaacharya, Monika Henzinger, Danupon Nanongkai, and Charalampos Tsourakakis. 15. Space-and ime-efficien algorihm for mainaining dense subgraphs on one-pass dynamic sreams. In STOC. ACM, [10] Moses Charikar. 00. Greedy approximaion algorihms for finding dense componens in a graph. In Approx. Algo. for Comb. Op [11] Jie Chen and Yousef Saad. 12. Dense subgraph exracion wih applicaion o communiy deecion. TKDE 24, 7 (12), [12] Michael S Crouch, Andrew McGregor, and Daniel Subbs. 13. Dynamic graphs in he sliding-window model. In ESA. Springer, [13] Mayur Daar, Arisides Gionis, Pior Indyk, and Rajeev Mowani. 02. Mainaining sream saisics over sliding windows. SIAM J. Compu. 31, 6 (02), [14] Yon Dourisboure, Filippo Geraci, and Marco Pellegrini. 07. Exracion and classificaion of dense communiies in he web. In WWW. ACM, [15] Alessandro Epaso, Silvio Laanzi, and Mauro Sozio. 15. Efficien Denses Subgraph Compuaion in Evolving Graphs. In WWW [16] David Eppsein, Zvi Galil, and Giuseppe F Ialiano Dynamic graph algorihms. Springer. [17] Esher Galbrun, Arisides Gionis, and Nikolaj Tai. 16. Top-k overlapping denses subgraphs. Daa Mining and Knowledge Discovery (16), [18] David Gibson, Ravi Kumar, and Andrew Tomkins. 05. Discovering large dense subgraphs in massive graphs. In VLDB. VLDB Endowmen, [19] Arisides Gionis and Charalampos E Tsourakakis. 15. Dense subgraph discovery: Kdd 15 uorial. In SIGKDD. ACM, [] Andrew V Goldberg Finding a maximum densiy subgraph. Universiy of California Berkeley, CA. [21] Guy Korsarz and David Peleg Generaing sparse 2-spanners. Journal of Algorihms 17, 2 (1994), [22] Jure Leskovec and Andrej Krevl. 14. SNAP Daases: Sanford Large Nework Daase Collecion. hp://snap.sanford.edu/daa. (June 14). [23] Rong-Hua Li, Jeffrey Xu Yu, and Rui Mao. 14. Efficien core mainenance in large dynamic graphs. TKDE 26, 10 (14), [24] Andrew McGregor, David Tench, Sofya Voronikova, and Hoa T Vu. 15. Denses Subgraph in Dynamic Graph Sreams. In MFCS. Springer, [25] James Orlin. 13. Max flows in O(nm) ime, or beer. In STOC [26] Polina Rozenshein, Aris Anagnosopoulos, Arisides Gionis, and Nikolaj Tai. 14. Even deecion in aciviy neworks. In SIGKDD [27] Ahme Erdem Saríyüce, Buğra Gedik, Gabriela Jacques-Silva, Kun-Lung Wu, and Ümi V Çaalyürek. 13. Sreaming algorihms for k-core decomposiion. VLDB 6, 6 (13), [28] Mauro Sozio and Arisides Gionis. 10. The communiy-search problem and how o plan a successful cockail pary. In KDD. ACM, [29] Isabelle Sanon and Gabriel Klio. 12. Sreaming graph pariioning for large disribued graphs. In KDD. ACM, [30] Nikolaj Tai and Arisides Gionis. 15. Densiy-friendly graph decomposiion. In WWW. ACM, [31] Charalampos Tsourakakis, Francesco Bonchi, Arisides Gionis, Francesco Gullo, and Maria Tsiarli. 13. Denser han he denses subgraph: exracing opimal quasi-cliques wih qualiy guaranees. In KDD. ACM, [32] Elena Valari, Maria Konaki, and Aposolos N Papadopoulos. 12. Discovery of op-k dense subgraphs in dynamic graph collecions. In SSDBM [33] Jaewon Yang and Jure Leskovec. 15. Defining and evaluaing nework communiies based on ground-ruh. ICML 42, 1 (15),

Implementing Ray Casting in Tetrahedral Meshes with Programmable Graphics Hardware (Technical Report)

Implementing Ray Casting in Tetrahedral Meshes with Programmable Graphics Hardware (Technical Report) Implemening Ray Casing in Terahedral Meshes wih Programmable Graphics Hardware (Technical Repor) Marin Kraus, Thomas Erl March 28, 2002 1 Inroducion Alhough cell-projecion, e.g., [3, 2], and resampling,

More information

Shortest Path Algorithms. Lecture I: Shortest Path Algorithms. Example. Graphs and Matrices. Setting: Dr Kieran T. Herley.

Shortest Path Algorithms. Lecture I: Shortest Path Algorithms. Example. Graphs and Matrices. Setting: Dr Kieran T. Herley. Shores Pah Algorihms Background Seing: Lecure I: Shores Pah Algorihms Dr Kieran T. Herle Deparmen of Compuer Science Universi College Cork Ocober 201 direced graph, real edge weighs Le he lengh of a pah

More information

Sam knows that his MP3 player has 40% of its battery life left and that the battery charges by an additional 12 percentage points every 15 minutes.

Sam knows that his MP3 player has 40% of its battery life left and that the battery charges by an additional 12 percentage points every 15 minutes. 8.F Baery Charging Task Sam wans o ake his MP3 player and his video game player on a car rip. An hour before hey plan o leave, he realized ha he forgo o charge he baeries las nigh. A ha poin, he plugged

More information

Image segmentation. Motivation. Objective. Definitions. A classification of segmentation techniques. Assumptions for thresholding

Image segmentation. Motivation. Objective. Definitions. A classification of segmentation techniques. Assumptions for thresholding Moivaion Image segmenaion Which pixels belong o he same objec in an image/video sequence? (spaial segmenaion) Which frames belong o he same video sho? (emporal segmenaion) Which frames belong o he same

More information

Coded Caching with Multiple File Requests

Coded Caching with Multiple File Requests Coded Caching wih Muliple File Requess Yi-Peng Wei Sennur Ulukus Deparmen of Elecrical and Compuer Engineering Universiy of Maryland College Park, MD 20742 ypwei@umd.edu ulukus@umd.edu Absrac We sudy a

More information

Optimal Crane Scheduling

Optimal Crane Scheduling Opimal Crane Scheduling Samid Hoda, John Hooker Laife Genc Kaya, Ben Peerson Carnegie Mellon Universiy Iiro Harjunkoski ABB Corporae Research EWO - 13 November 2007 1/16 Problem Track-mouned cranes move

More information

Learning in Games via Opponent Strategy Estimation and Policy Search

Learning in Games via Opponent Strategy Estimation and Policy Search Learning in Games via Opponen Sraegy Esimaion and Policy Search Yavar Naddaf Deparmen of Compuer Science Universiy of Briish Columbia Vancouver, BC yavar@naddaf.name Nando de Freias (Supervisor) Deparmen

More information

COSC 3213: Computer Networks I Chapter 6 Handout # 7

COSC 3213: Computer Networks I Chapter 6 Handout # 7 COSC 3213: Compuer Neworks I Chaper 6 Handou # 7 Insrucor: Dr. Marvin Mandelbaum Deparmen of Compuer Science York Universiy F05 Secion A Medium Access Conrol (MAC) Topics: 1. Muliple Access Communicaions:

More information

An Efficient Delivery Scheme for Coded Caching

An Efficient Delivery Scheme for Coded Caching 201 27h Inernaional Teleraffic Congress An Efficien Delivery Scheme for Coded Caching Abinesh Ramakrishnan, Cedric Wesphal and Ahina Markopoulou Deparmen of Elecrical Engineering and Compuer Science, Universiy

More information

EECS 487: Interactive Computer Graphics

EECS 487: Interactive Computer Graphics EECS 487: Ineracive Compuer Graphics Lecure 7: B-splines curves Raional Bézier and NURBS Cubic Splines A represenaion of cubic spline consiss of: four conrol poins (why four?) hese are compleely user specified

More information

The Impact of Product Development on the Lifecycle of Defects

The Impact of Product Development on the Lifecycle of Defects The Impac of Produc Developmen on he Lifecycle of Rudolf Ramler Sofware Compeence Cener Hagenberg Sofware Park 21 A-4232 Hagenberg, Ausria +43 7236 3343 872 rudolf.ramler@scch.a ABSTRACT This paper invesigaes

More information

Scheduling. Scheduling. EDA421/DIT171 - Parallel and Distributed Real-Time Systems, Chalmers/GU, 2011/2012 Lecture #4 Updated March 16, 2012

Scheduling. Scheduling. EDA421/DIT171 - Parallel and Distributed Real-Time Systems, Chalmers/GU, 2011/2012 Lecture #4 Updated March 16, 2012 EDA421/DIT171 - Parallel and Disribued Real-Time Sysems, Chalmers/GU, 2011/2012 Lecure #4 Updaed March 16, 2012 Aemps o mee applicaion consrains should be done in a proacive way hrough scheduling. Schedule

More information

Data Structures and Algorithms. The material for this lecture is drawn, in part, from The Practice of Programming (Kernighan & Pike) Chapter 2

Data Structures and Algorithms. The material for this lecture is drawn, in part, from The Practice of Programming (Kernighan & Pike) Chapter 2 Daa Srucures and Algorihms The maerial for his lecure is drawn, in par, from The Pracice of Programming (Kernighan & Pike) Chaper 2 1 Moivaing Quoaion Every program depends on algorihms and daa srucures,

More information

NRMI: Natural and Efficient Middleware

NRMI: Natural and Efficient Middleware NRMI: Naural and Efficien Middleware Eli Tilevich and Yannis Smaragdakis Cener for Experimenal Research in Compuer Sysems (CERCS), College of Compuing, Georgia Tech {ilevich, yannis}@cc.gaech.edu Absrac

More information

A Matching Algorithm for Content-Based Image Retrieval

A Matching Algorithm for Content-Based Image Retrieval A Maching Algorihm for Conen-Based Image Rerieval Sue J. Cho Deparmen of Compuer Science Seoul Naional Universiy Seoul, Korea Absrac Conen-based image rerieval sysem rerieves an image from a daabase using

More information

Assignment 2. Due Monday Feb. 12, 10:00pm.

Assignment 2. Due Monday Feb. 12, 10:00pm. Faculy of rs and Science Universiy of Torono CSC 358 - Inroducion o Compuer Neworks, Winer 218, LEC11 ssignmen 2 Due Monday Feb. 12, 1:pm. 1 Quesion 1 (2 Poins): Go-ack n RQ In his quesion, we review how

More information

4 Error Control. 4.1 Issues with Reliable Protocols

4 Error Control. 4.1 Issues with Reliable Protocols 4 Error Conrol Jus abou all communicaion sysems aemp o ensure ha he daa ges o he oher end of he link wihou errors. Since i s impossible o build an error-free physical layer (alhough some shor links can

More information

Lecture 18: Mix net Voting Systems

Lecture 18: Mix net Voting Systems 6.897: Advanced Topics in Crypography Apr 9, 2004 Lecure 18: Mix ne Voing Sysems Scribed by: Yael Tauman Kalai 1 Inroducion In he previous lecure, we defined he noion of an elecronic voing sysem, and specified

More information

CAMERA CALIBRATION BY REGISTRATION STEREO RECONSTRUCTION TO 3D MODEL

CAMERA CALIBRATION BY REGISTRATION STEREO RECONSTRUCTION TO 3D MODEL CAMERA CALIBRATION BY REGISTRATION STEREO RECONSTRUCTION TO 3D MODEL Klečka Jan Docoral Degree Programme (1), FEEC BUT E-mail: xkleck01@sud.feec.vubr.cz Supervised by: Horák Karel E-mail: horak@feec.vubr.cz

More information

A time-space consistency solution for hardware-in-the-loop simulation system

A time-space consistency solution for hardware-in-the-loop simulation system Inernaional Conference on Advanced Elecronic Science and Technology (AEST 206) A ime-space consisency soluion for hardware-in-he-loop simulaion sysem Zexin Jiang a Elecric Power Research Insiue of Guangdong

More information

Chapter 8 LOCATION SERVICES

Chapter 8 LOCATION SERVICES Disribued Compuing Group Chaper 8 LOCATION SERVICES Mobile Compuing Winer 2005 / 2006 Overview Mobile IP Moivaion Daa ransfer Encapsulaion Locaion Services & Rouing Classificaion of locaion services Home

More information

Image Content Representation

Image Content Representation Image Conen Represenaion Represenaion for curves and shapes regions relaionships beween regions E.G.M. Perakis Image Represenaion & Recogniion 1 Reliable Represenaion Uniqueness: mus uniquely specify an

More information

Less Pessimistic Worst-Case Delay Analysis for Packet-Switched Networks

Less Pessimistic Worst-Case Delay Analysis for Packet-Switched Networks Less Pessimisic Wors-Case Delay Analysis for Packe-Swiched Neworks Maias Wecksén Cenre for Research on Embedded Sysems P O Box 823 SE-31 18 Halmsad maias.wecksen@hh.se Magnus Jonsson Cenre for Research

More information

An Adaptive Spatial Depth Filter for 3D Rendering IP

An Adaptive Spatial Depth Filter for 3D Rendering IP JOURNAL OF SEMICONDUCTOR TECHNOLOGY AND SCIENCE, VOL.3, NO. 4, DECEMBER, 23 175 An Adapive Spaial Deph Filer for 3D Rendering IP Chang-Hyo Yu and Lee-Sup Kim Absrac In his paper, we presen a new mehod

More information

Analysis of Various Types of Bugs in the Object Oriented Java Script Language Coding

Analysis of Various Types of Bugs in the Object Oriented Java Script Language Coding Indian Journal of Science and Technology, Vol 8(21), DOI: 10.17485/ijs/2015/v8i21/69958, Sepember 2015 ISSN (Prin) : 0974-6846 ISSN (Online) : 0974-5645 Analysis of Various Types of Bugs in he Objec Oriened

More information

Partition-based document identifier assignment (PBDIA) algorithm. (long queries)

Partition-based document identifier assignment (PBDIA) algorithm. (long queries) ( ) Pariion-based documen idenifier assignmen (PBDIA) algorihm PBDIA (long queries) (parallel IR) :,,,, d-gap Compressing an invered file can grealy improve query performance of an informaion rerieval

More information

Real Time Contextual Summarization of Highly Dynamic Data Streams

Real Time Contextual Summarization of Highly Dynamic Data Streams Real Time Conexual Summarizaion of Highly Dynamic Daa Sreams Manoj K Agarwal Microsof Bing Search Technology Cener - India Hyderabad 500032, India agarwalm@microsof.com Krihi Ramamriham Dep. of Compuer

More information

Constant-Work-Space Algorithms for Shortest Paths in Trees and Simple Polygons

Constant-Work-Space Algorithms for Shortest Paths in Trees and Simple Polygons Journal of Graph Algorihms and Applicaions hp://jgaa.info/ vol. 15, no. 5, pp. 569 586 (2011) Consan-Work-Space Algorihms for Shores Pahs in Trees and Simple Polygons Tesuo Asano 1 Wolfgang Mulzer 2 Yajun

More information

MATH Differential Equations September 15, 2008 Project 1, Fall 2008 Due: September 24, 2008

MATH Differential Equations September 15, 2008 Project 1, Fall 2008 Due: September 24, 2008 MATH 5 - Differenial Equaions Sepember 15, 8 Projec 1, Fall 8 Due: Sepember 4, 8 Lab 1.3 - Logisics Populaion Models wih Harvesing For his projec we consider lab 1.3 of Differenial Equaions pages 146 o

More information

Network management and QoS provisioning - QoS in Frame Relay. . packet switching with virtual circuit service (virtual circuits are bidirectional);

Network management and QoS provisioning - QoS in Frame Relay. . packet switching with virtual circuit service (virtual circuits are bidirectional); QoS in Frame Relay Frame relay characerisics are:. packe swiching wih virual circui service (virual circuis are bidirecional);. labels are called DLCI (Daa Link Connecion Idenifier);. for connecion is

More information

Packet Scheduling in a Low-Latency Optical Interconnect with Electronic Buffers

Packet Scheduling in a Low-Latency Optical Interconnect with Electronic Buffers Packe cheduling in a Low-Laency Opical Inerconnec wih Elecronic Buffers Lin Liu Zhenghao Zhang Yuanyuan Yang Dep Elecrical & Compuer Engineering Compuer cience Deparmen Dep Elecrical & Compuer Engineering

More information

STEREO PLANE MATCHING TECHNIQUE

STEREO PLANE MATCHING TECHNIQUE STEREO PLANE MATCHING TECHNIQUE Commission III KEY WORDS: Sereo Maching, Surface Modeling, Projecive Transformaion, Homography ABSTRACT: This paper presens a new ype of sereo maching algorihm called Sereo

More information

Quantitative macro models feature an infinite number of periods A more realistic (?) view of time

Quantitative macro models feature an infinite number of periods A more realistic (?) view of time INFINIE-HORIZON CONSUMPION-SAVINGS MODEL SEPEMBER, Inroducion BASICS Quaniaive macro models feaure an infinie number of periods A more realisic (?) view of ime Infinie number of periods A meaphor for many

More information

4. Minimax and planning problems

4. Minimax and planning problems CS/ECE/ISyE 524 Inroducion o Opimizaion Spring 2017 18 4. Minima and planning problems ˆ Opimizing piecewise linear funcions ˆ Minima problems ˆ Eample: Chebyshev cener ˆ Muli-period planning problems

More information

Computer representations of piecewise

Computer representations of piecewise Edior: Gabriel Taubin Inroducion o Geomeric Processing hrough Opimizaion Gabriel Taubin Brown Universiy Compuer represenaions o piecewise smooh suraces have become vial echnologies in areas ranging rom

More information

Nonparametric CUSUM Charts for Process Variability

Nonparametric CUSUM Charts for Process Variability Journal of Academia and Indusrial Research (JAIR) Volume 3, Issue June 4 53 REEARCH ARTICLE IN: 78-53 Nonparameric CUUM Chars for Process Variabiliy D.M. Zombade and V.B. Ghue * Dep. of aisics, Walchand

More information

Outline. CS38 Introduction to Algorithms 5/8/2014. Network flow. Lecture 12 May 8, 2014

Outline. CS38 Introduction to Algorithms 5/8/2014. Network flow. Lecture 12 May 8, 2014 /8/0 Ouline CS8 Inroducion o Algorihm Lecure May 8, 0 Nework flow finihing capaciy-caling analyi Edmond-Karp, blocking-flow implemenaion uni-capaciy imple graph biparie maching edge-dijoin pah aignmen

More information

Rule-Based Multi-Query Optimization

Rule-Based Multi-Query Optimization Rule-Based Muli-Query Opimizaion Mingsheng Hong Dep. of Compuer cience Cornell Universiy mshong@cs.cornell.edu Johannes Gehrke Dep. of Compuer cience Cornell Universiy johannes@cs.cornell.edu Mirek Riedewald

More information

MORPHOLOGICAL SEGMENTATION OF IMAGE SEQUENCES

MORPHOLOGICAL SEGMENTATION OF IMAGE SEQUENCES MORPHOLOGICAL SEGMENTATION OF IMAGE SEQUENCES B. MARCOTEGUI and F. MEYER Ecole des Mines de Paris, Cenre de Morphologie Mahémaique, 35, rue Sain-Honoré, F 77305 Fonainebleau Cedex, France Absrac. In image

More information

The Data Locality of Work Stealing

The Data Locality of Work Stealing The Daa Localiy of Work Sealing Umu A. Acar School of Compuer Science Carnegie Mellon Universiy umu@cs.cmu.edu Guy E. Blelloch School of Compuer Science Carnegie Mellon Universiy guyb@cs.cmu.edu Rober

More information

Restorable Dynamic Quality of Service Routing

Restorable Dynamic Quality of Service Routing QOS ROUTING Resorable Dynamic Qualiy of Service Rouing Murali Kodialam and T. V. Lakshman, Lucen Technologies ABSTRACT The focus of qualiy-of-service rouing has been on he rouing of a single pah saisfying

More information

Performance Evaluation of Implementing Calls Prioritization with Different Queuing Disciplines in Mobile Wireless Networks

Performance Evaluation of Implementing Calls Prioritization with Different Queuing Disciplines in Mobile Wireless Networks Journal of Compuer Science 2 (5): 466-472, 2006 ISSN 1549-3636 2006 Science Publicaions Performance Evaluaion of Implemening Calls Prioriizaion wih Differen Queuing Disciplines in Mobile Wireless Neworks

More information

Scalable Big Graph Processing in MapReduce

Scalable Big Graph Processing in MapReduce Scalable Big Graph Processing in MapReduce Lu Qin,JeffreyXuYu,LijunChang,HongCheng,ChengqiZhang, Xuemin Lin Cenre for Quanum Compuaion and Inelligen Sysems, Universiy of Technology, Sydney, Ausralia The

More information

Improving the Efficiency of Dynamic Service Provisioning in Transport Networks with Scheduled Services

Improving the Efficiency of Dynamic Service Provisioning in Transport Networks with Scheduled Services Improving he Efficiency of Dynamic Service Provisioning in Transpor Neworks wih Scheduled Services Ralf Hülsermann, Monika Jäger and Andreas Gladisch Technologiezenrum, T-Sysems, Goslarer Ufer 35, D-1585

More information

Simple Network Management Based on PHP and SNMP

Simple Network Management Based on PHP and SNMP Simple Nework Managemen Based on PHP and SNMP Krasimir Trichkov, Elisavea Trichkova bsrac: This paper aims o presen simple mehod for nework managemen based on SNMP - managemen of Cisco rouer. The paper

More information

Difficulty-aware Hybrid Search in Peer-to-Peer Networks

Difficulty-aware Hybrid Search in Peer-to-Peer Networks Difficuly-aware Hybrid Search in Peer-o-Peer Neworks Hanhua Chen, Hai Jin, Yunhao Liu, Lionel M. Ni School of Compuer Science and Technology Huazhong Univ. of Science and Technology {chenhanhua, hjin}@hus.edu.cn

More information

Parallel and Distributed Systems for Constructive Neural Network Learning*

Parallel and Distributed Systems for Constructive Neural Network Learning* Parallel and Disribued Sysems for Consrucive Neural Nework Learning* J. Flecher Z. Obradovi School of Elecrical Engineering and Compuer Science Washingon Sae Universiy Pullman WA 99164-2752 Absrac A consrucive

More information

Spline Curves. Color Interpolation. Normal Interpolation. Last Time? Today. glshademodel (GL_SMOOTH); Adjacency Data Structures. Mesh Simplification

Spline Curves. Color Interpolation. Normal Interpolation. Last Time? Today. glshademodel (GL_SMOOTH); Adjacency Data Structures. Mesh Simplification Las Time? Adjacency Daa Srucures Spline Curves Geomeric & opologic informaion Dynamic allocaion Efficiency of access Mesh Simplificaion edge collapse/verex spli geomorphs progressive ransmission view-dependen

More information

Design Alternatives for a Thin Lens Spatial Integrator Array

Design Alternatives for a Thin Lens Spatial Integrator Array Egyp. J. Solids, Vol. (7), No. (), (004) 75 Design Alernaives for a Thin Lens Spaial Inegraor Array Hala Kamal *, Daniel V azquez and Javier Alda and E. Bernabeu Opics Deparmen. Universiy Compluense of

More information

The Roots of Lisp paul graham

The Roots of Lisp paul graham The Roos of Lisp paul graham Draf, January 18, 2002. In 1960, John McCarhy published a remarkable paper in which he did for programming somehing like wha Euclid did for geomery. 1 He showed how, given

More information

NEWTON S SECOND LAW OF MOTION

NEWTON S SECOND LAW OF MOTION Course and Secion Dae Names NEWTON S SECOND LAW OF MOTION The acceleraion of an objec is defined as he rae of change of elociy. If he elociy changes by an amoun in a ime, hen he aerage acceleraion during

More information

Voltair Version 2.5 Release Notes (January, 2018)

Voltair Version 2.5 Release Notes (January, 2018) Volair Version 2.5 Release Noes (January, 2018) Inroducion 25-Seven s new Firmware Updae 2.5 for he Volair processor is par of our coninuing effors o improve Volair wih new feaures and capabiliies. For

More information

Web System for the Remote Control and Execution of an IEC Application

Web System for the Remote Control and Execution of an IEC Application Web Sysem for he Remoe Conrol and Execuion of an IEC 61499 Applicaion Oana ROHAT, Dan POPESCU Faculy of Auomaion and Compuer Science, Poliehnica Universiy, Splaiul Independenței 313, Bucureși, 060042,

More information

Michiel Helder and Marielle C.T.A Geurts. Hoofdkantoor PTT Post / Dutch Postal Services Headquarters

Michiel Helder and Marielle C.T.A Geurts. Hoofdkantoor PTT Post / Dutch Postal Services Headquarters SHORT TERM PREDICTIONS A MONITORING SYSTEM by Michiel Helder and Marielle C.T.A Geurs Hoofdkanoor PTT Pos / Duch Posal Services Headquarers Keywords macro ime series shor erm predicions ARIMA-models faciliy

More information

Quick Verification of Concurrent Programs by Iteratively Relaxed Scheduling

Quick Verification of Concurrent Programs by Iteratively Relaxed Scheduling Quick Verificaion of Concurren Programs by Ieraively Relaxed Scheduling Parick Mezler, Habib Saissi, Péer Bokor, Neeraj Suri Technische Univerisä Darmsad, Germany {mezler, saissi, pbokor, suri}@deeds.informaik.u-darmsad.de

More information

Opportunistic Flooding in Low-Duty-Cycle Wireless Sensor Networks with Unreliable Links

Opportunistic Flooding in Low-Duty-Cycle Wireless Sensor Networks with Unreliable Links 1 in Low-uy-ycle Wireless Sensor Neworks wih Unreliable Links Shuo uo, Suden Member, IEEE, Liang He, Member, IEEE, Yu u, Member, IEEE, o Jiang, Suden Member, IEEE, and Tian He, Member, IEEE bsrac looding

More information

Landmarks: A New Model for Similarity-Based Pattern Querying in Time Series Databases

Landmarks: A New Model for Similarity-Based Pattern Querying in Time Series Databases Lmarks: A New Model for Similariy-Based Paern Querying in Time Series Daabases Chang-Shing Perng Haixun Wang Sylvia R. Zhang D. So Parker perng@cs.ucla.edu hxwang@cs.ucla.edu Sylvia Zhang@cle.com so@cs.ucla.edu

More information

Adaptive Workflow Scheduling on Cloud Computing Platforms with Iterative Ordinal Optimization

Adaptive Workflow Scheduling on Cloud Computing Platforms with Iterative Ordinal Optimization Adapive Workflow Scheduling on Cloud Compuing Plaforms wih Ieraive Ordinal Opimizaion Fan Zhang, Senior Member, IEEE; Junwei Cao, Senior Member, IEEE; Kai Hwang, Fellow, IEEE; Keqin Li, Senior Member,

More information

BI-TEMPORAL INDEXING

BI-TEMPORAL INDEXING BI-TEMPORAL INDEXING Mirella M. Moro Uniersidade Federal do Rio Grande do Sul Poro Alegre, RS, Brazil hp://www.inf.ufrgs.br/~mirella/ Vassilis J. Tsoras Uniersiy of California, Rierside Rierside, CA 92521,

More information

A Tool for Multi-Hour ATM Network Design considering Mixed Peer-to-Peer and Client-Server based Services

A Tool for Multi-Hour ATM Network Design considering Mixed Peer-to-Peer and Client-Server based Services A Tool for Muli-Hour ATM Nework Design considering Mied Peer-o-Peer and Clien-Server based Services Conac Auhor Name: Luis Cardoso Company / Organizaion: Porugal Telecom Inovação Complee Mailing Address:

More information

Automatic Calculation of Coverage Profiles for Coverage-based Testing

Automatic Calculation of Coverage Profiles for Coverage-based Testing Auomaic Calculaion of Coverage Profiles for Coverage-based Tesing Raimund Kirner 1 and Waler Haas 1 Vienna Universiy of Technology, Insiue of Compuer Engineering, Vienna, Ausria, raimund@vmars.uwien.ac.a

More information

Precise Voronoi Cell Extraction of Free-form Rational Planar Closed Curves

Precise Voronoi Cell Extraction of Free-form Rational Planar Closed Curves Precise Voronoi Cell Exracion of Free-form Raional Planar Closed Curves Iddo Hanniel, Ramanahan Muhuganapahy, Gershon Elber Deparmen of Compuer Science Technion, Israel Insiue of Technology Haifa 32000,

More information

A Routing Algorithm for Flip-Chip Design

A Routing Algorithm for Flip-Chip Design A Rouing Algorihm for Flip-hip Design Jia-Wei Fang, I-Jye Lin, and Yao-Wen hang, Graduae Insiue of Elecronics Engineering, Naional Taiwan Universiy, Taipei Deparmen of Elecrical Engineering, Naional Taiwan

More information

Motor Control. 5. Control. Motor Control. Motor Control

Motor Control. 5. Control. Motor Control. Motor Control 5. Conrol In his chaper we will do: Feedback Conrol On/Off Conroller PID Conroller Moor Conrol Why use conrol a all? Correc or wrong? Supplying a cerain volage / pulsewidh will make he moor spin a a cerain

More information

Gauss-Jordan Algorithm

Gauss-Jordan Algorithm Gauss-Jordan Algorihm The Gauss-Jordan algorihm is a sep by sep procedure for solving a sysem of linear equaions which may conain any number of variables and any number of equaions. The algorihm is carried

More information

Handling uncertainty in semantic information retrieval process

Handling uncertainty in semantic information retrieval process Handling uncerainy in semanic informaion rerieval process Chkiwa Mounira 1, Jedidi Anis 1 and Faiez Gargouri 1 1 Mulimedia, InfoRmaion sysems and Advanced Compuing Laboraory Sfax Universiy, Tunisia m.chkiwa@gmail.com,

More information

Mobile Robots Mapping

Mobile Robots Mapping Mobile Robos Mapping 1 Roboics is Easy conrol behavior percepion modelling domain model environmen model informaion exracion raw daa planning ask cogniion reasoning pah planning navigaion pah execuion

More information

COMP26120: Algorithms and Imperative Programming

COMP26120: Algorithms and Imperative Programming COMP26120 ecure C3 1/48 COMP26120: Algorihms and Imperaive Programming ecure C3: C - Recursive Daa Srucures Pee Jinks School of Compuer Science, Universiy of Mancheser Auumn 2011 COMP26120 ecure C3 2/48

More information

IntentSearch:Capturing User Intention for One-Click Internet Image Search

IntentSearch:Capturing User Intention for One-Click Internet Image Search JOURNAL OF L A T E X CLASS FILES, VOL. 6, NO. 1, JANUARY 2010 1 InenSearch:Capuring User Inenion for One-Click Inerne Image Search Xiaoou Tang, Fellow, IEEE, Ke Liu, Jingyu Cui, Suden Member, IEEE, Fang

More information

Reinforcement Learning by Policy Improvement. Making Use of Experiences of The Other Tasks. Hajime Kimura and Shigenobu Kobayashi

Reinforcement Learning by Policy Improvement. Making Use of Experiences of The Other Tasks. Hajime Kimura and Shigenobu Kobayashi Reinforcemen Learning by Policy Improvemen Making Use of Experiences of The Oher Tasks Hajime Kimura and Shigenobu Kobayashi Tokyo Insiue of Technology, JAPAN genfe.dis.iech.ac.jp, kobayasidis.iech.ac.jp

More information

An Improved Square-Root Nyquist Shaping Filter

An Improved Square-Root Nyquist Shaping Filter An Improved Square-Roo Nyquis Shaping Filer fred harris San Diego Sae Universiy fred.harris@sdsu.edu Sridhar Seshagiri San Diego Sae Universiy Seshigar.@engineering.sdsu.edu Chris Dick Xilinx Corp. chris.dick@xilinx.com

More information

A Progressive-ILP Based Routing Algorithm for Cross-Referencing Biochips

A Progressive-ILP Based Routing Algorithm for Cross-Referencing Biochips 16.3 A Progressive-ILP Based Rouing Algorihm for Cross-Referencing Biochips Ping-Hung Yuh 1, Sachin Sapanekar 2, Chia-Lin Yang 1, Yao-Wen Chang 3 1 Deparmen of Compuer Science and Informaion Engineering,

More information

Video-Based Face Recognition Using Probabilistic Appearance Manifolds

Video-Based Face Recognition Using Probabilistic Appearance Manifolds Video-Based Face Recogniion Using Probabilisic Appearance Manifolds Kuang-Chih Lee Jeffrey Ho Ming-Hsuan Yang David Kriegman klee10@uiuc.edu jho@cs.ucsd.edu myang@honda-ri.com kriegman@cs.ucsd.edu Compuer

More information

CENG 477 Introduction to Computer Graphics. Modeling Transformations

CENG 477 Introduction to Computer Graphics. Modeling Transformations CENG 477 Inroducion o Compuer Graphics Modeling Transformaions Modeling Transformaions Model coordinaes o World coordinaes: Model coordinaes: All shapes wih heir local coordinaes and sies. world World

More information

PART 1 REFERENCE INFORMATION CONTROL DATA 6400 SYSTEMS CENTRAL PROCESSOR MONITOR

PART 1 REFERENCE INFORMATION CONTROL DATA 6400 SYSTEMS CENTRAL PROCESSOR MONITOR . ~ PART 1 c 0 \,).,,.,, REFERENCE NFORMATON CONTROL DATA 6400 SYSTEMS CENTRAL PROCESSOR MONTOR n CONTROL DATA 6400 Compuer Sysems, sysem funcions are normally handled by he Monior locaed in a Peripheral

More information

Audio Engineering Society. Convention Paper. Presented at the 119th Convention 2005 October 7 10 New York, New York USA

Audio Engineering Society. Convention Paper. Presented at the 119th Convention 2005 October 7 10 New York, New York USA Audio Engineering Sociey Convenion Paper Presened a he 119h Convenion 2005 Ocober 7 10 New Yor, New Yor USA This convenion paper has been reproduced from he auhor's advance manuscrip, wihou ediing, correcions,

More information

SEINA: A Stealthy and Effective Internal Attack in Hadoop Systems

SEINA: A Stealthy and Effective Internal Attack in Hadoop Systems SEINA: A Sealhy and Effecive Inernal Aack in Hadoop Sysems Jiayin Wang, Teng Wang, Zhengyu Yang, Ying ao, Ningfang i, and Bo Sheng Deparmen of Compuer Science, Universiy of assachuses Boson, 1 orrissey

More information

Rao-Blackwellized Particle Filtering for Probing-Based 6-DOF Localization in Robotic Assembly

Rao-Blackwellized Particle Filtering for Probing-Based 6-DOF Localization in Robotic Assembly MITSUBISHI ELECTRIC RESEARCH LABORATORIES hp://www.merl.com Rao-Blackwellized Paricle Filering for Probing-Based 6-DOF Localizaion in Roboic Assembly Yuichi Taguchi, Tim Marks, Haruhisa Okuda TR1-8 June

More information

Time Expression Recognition Using a Constituent-based Tagging Scheme

Time Expression Recognition Using a Constituent-based Tagging Scheme Track: Web Conen Analysis, Semanics and Knowledge Time Expression Recogniion Using a Consiuen-based Tagging Scheme Xiaoshi Zhong and Erik Cambria School of Compuer Science and Engineering Nanyang Technological

More information

Y. Tsiatouhas. VLSI Systems and Computer Architecture Lab

Y. Tsiatouhas. VLSI Systems and Computer Architecture Lab CMOS INEGRAED CIRCUI DESIGN ECHNIQUES Universiy of Ioannina Clocking Schemes Dep. of Compuer Science and Engineering Y. siaouhas CMOS Inegraed Circui Design echniques Overview 1. Jier Skew hroughpu Laency

More information

Improving Ranking of Search Engines Results Based on Power Links

Improving Ranking of Search Engines Results Based on Power Links IPASJ Inernaional Journal of Informaion Technology (IIJIT) Web Sie: hp://www.ipasj.org/iijit/iijit.hm A Publisher for Research Moivaion... Email: edioriiji@ipasj.org Volume 2, Issue 9, Sepember 2014 ISSN

More information

Test - Accredited Configuration Engineer (ACE) Exam - PAN-OS 6.0 Version

Test - Accredited Configuration Engineer (ACE) Exam - PAN-OS 6.0 Version Tes - Accredied Configuraion Engineer (ACE) Exam - PAN-OS 6.0 Version ACE Exam Quesion 1 of 50. Which of he following saemens is NOT abou Palo Alo Neworks firewalls? Sysem defauls may be resored by performing

More information

Visual Indoor Localization with a Floor-Plan Map

Visual Indoor Localization with a Floor-Plan Map Visual Indoor Localizaion wih a Floor-Plan Map Hang Chu Dep. of ECE Cornell Universiy Ihaca, NY 14850 hc772@cornell.edu Absrac In his repor, a indoor localizaion mehod is presened. The mehod akes firsperson

More information

Utility-Based Hybrid Memory Management

Utility-Based Hybrid Memory Management Uiliy-Based Hybrid Memory Managemen Yang Li Saugaa Ghose Jongmoo Choi Jin Sun Hui Wang Onur Mulu Carnegie Mellon Universiy Dankook Universiy Beihang Universiy ETH Zürich While he memory fooprins of cloud

More information

Dynamic Route Planning and Obstacle Avoidance Model for Unmanned Aerial Vehicles

Dynamic Route Planning and Obstacle Avoidance Model for Unmanned Aerial Vehicles Volume 116 No. 24 2017, 315-329 ISSN: 1311-8080 (prined version); ISSN: 1314-3395 (on-line version) url: hp://www.ijpam.eu ijpam.eu Dynamic Roue Planning and Obsacle Avoidance Model for Unmanned Aerial

More information

M(t)/M/1 Queueing System with Sinusoidal Arrival Rate

M(t)/M/1 Queueing System with Sinusoidal Arrival Rate 20 TUTA/IOE/PCU Journal of he Insiue of Engineering, 205, (): 20-27 TUTA/IOE/PCU Prined in Nepal M()/M/ Queueing Sysem wih Sinusoidal Arrival Rae A.P. Pan, R.P. Ghimire 2 Deparmen of Mahemaics, Tri-Chandra

More information

In fmri a Dual Echo Time EPI Pulse Sequence Can Induce Sources of Error in Dynamic Magnetic Field Maps

In fmri a Dual Echo Time EPI Pulse Sequence Can Induce Sources of Error in Dynamic Magnetic Field Maps In fmri a Dual Echo Time EPI Pulse Sequence Can Induce Sources of Error in Dynamic Magneic Field Maps A. D. Hahn 1, A. S. Nencka 1 and D. B. Rowe 2,1 1 Medical College of Wisconsin, Milwaukee, WI, Unied

More information

A GRAPHICS PROCESSING UNIT IMPLEMENTATION OF THE PARTICLE FILTER

A GRAPHICS PROCESSING UNIT IMPLEMENTATION OF THE PARTICLE FILTER A GRAPHICS PROCESSING UNIT IMPLEMENTATION OF THE PARTICLE FILTER ABSTRACT Modern graphics cards for compuers, and especially heir graphics processing unis (GPUs), are designed for fas rendering of graphics.

More information

Locating Internet Bottlenecks: Algorithms, Measurements, and Implications

Locating Internet Bottlenecks: Algorithms, Measurements, and Implications Locaing Inerne Bolenecks: Algorihms, Measuremens, and Implicaions Ningning Hu Li (Erran) Li Zhuoing Morley Mao Carnegie Mellon Universiy Bell Laboraories Universiy of Michigan hnn@cs.cmu.edu erranlli@bell-labs.com

More information

source managemen, naming, proecion, and service provisions. This paper concenraes on he basic processor scheduling aspecs of resource managemen. 2 The

source managemen, naming, proecion, and service provisions. This paper concenraes on he basic processor scheduling aspecs of resource managemen. 2 The Virual Compuers A New Paradigm for Disribued Operaing Sysems Banu Ozden y Aaron J. Goldberg Avi Silberschaz z 600 Mounain Ave. AT&T Bell Laboraories Murray Hill, NJ 07974 Absrac The virual compuers (VC)

More information

Chapter 3 MEDIA ACCESS CONTROL

Chapter 3 MEDIA ACCESS CONTROL Chaper 3 MEDIA ACCESS CONTROL Overview Moivaion SDMA, FDMA, TDMA Aloha Adapive Aloha Backoff proocols Reservaion schemes Polling Disribued Compuing Group Mobile Compuing Summer 2003 Disribued Compuing

More information

SOT: Compact Representation for Triangle and Tetrahedral Meshes

SOT: Compact Representation for Triangle and Tetrahedral Meshes SOT: Compac Represenaion for Triangle and Terahedral Meshes Topraj Gurung and Jarek Rossignac School of Ineracive Compuing, College of Compuing, Georgia Insiue of Technology, Alana, GA ABSTRACT The Corner

More information

Data Structures and Algorithms

Data Structures and Algorithms Daa Srucures and Algorihms The maerial for his lecure is drawn, in ar, from The Pracice of Programming (Kernighan & Pike) Chaer 2 1 Goals of his Lecure Hel you learn (or refresh your memory) abou: Common

More information

Why not experiment with the system itself? Ways to study a system System. Application areas. Different kinds of systems

Why not experiment with the system itself? Ways to study a system System. Application areas. Different kinds of systems Simulaion Wha is simulaion? Simple synonym: imiaion We are ineresed in sudying a Insead of experimening wih he iself we experimen wih a model of he Experimen wih he Acual Ways o sudy a Sysem Experimen

More information

The Beer Dock: Three and a Half Implementations of the Beer Distribution Game

The Beer Dock: Three and a Half Implementations of the Beer Distribution Game The Beer Dock 2002-08-13 17:55:44-0700 The Beer Dock: Three and a Half Implemenaions of he Beer Disribuion Game Michael J. Norh[1] and Charles M. Macal Argonne Naional Laboraory, Argonne, Illinois Absrac

More information

Evaluation and Improvement of Region-based Motion Segmentation

Evaluation and Improvement of Region-based Motion Segmentation Evaluaion and Improvemen of Region-based Moion Segmenaion Mark Ross Universiy Koblenz-Landau, Insiue of Compuaional Visualisics, Universiässraße 1, 56070 Koblenz, Germany Email: ross@uni-koblenz.de Absrac

More information

Real Time Integral-Based Structural Health Monitoring

Real Time Integral-Based Structural Health Monitoring Real Time Inegral-Based Srucural Healh Monioring The nd Inernaional Conference on Sensing Technology ICST 7 J. G. Chase, I. Singh-Leve, C. E. Hann, X. Chen Deparmen of Mechanical Engineering, Universiy

More information

Video Content Description Using Fuzzy Spatio-Temporal Relations

Video Content Description Using Fuzzy Spatio-Temporal Relations Proceedings of he 4s Hawaii Inernaional Conference on Sysem Sciences - 008 Video Conen Descripion Using Fuzzy Spaio-Temporal Relaions rchana M. Rajurkar *, R.C. Joshi and Sananu Chaudhary 3 Dep of Compuer

More information

WITH the basic principles and algorithms of structurefrom-motion

WITH the basic principles and algorithms of structurefrom-motion JOURNAL OF L A T E X CLASS FILES, VOL., NO. 8, AUGUST 5 Coreses for Triangulaion Qianggong Zhang and Ta-Jun Chin arxiv:77.566v [cs.cg] 8 Jul 7 Absrac Muliple-view riangulaion by l minimisaion has become

More information

A Principled Approach to. MILP Modeling. Columbia University, August Carnegie Mellon University. Workshop on MIP. John Hooker.

A Principled Approach to. MILP Modeling. Columbia University, August Carnegie Mellon University. Workshop on MIP. John Hooker. Slide A Principled Approach o MILP Modeling John Hooer Carnegie Mellon Universiy Worshop on MIP Columbia Universiy, Augus 008 Proposal MILP modeling is an ar, bu i need no be unprincipled. Slide Proposal

More information