STABILITY OF SIMULTANEOUS RECURRENT NEURAL NETWORK DYNAMICS FOR STATIC OPTIMIZATION

Size: px
Start display at page:

Download "STABILITY OF SIMULTANEOUS RECURRENT NEURAL NETWORK DYNAMICS FOR STATIC OPTIMIZATION"

Transcription

1 STABILITY OF SIMULTANEOUS RECURRENT NEURAL NETWOR DYNAMICS FOR STATIC OPTIMIZATION Grsel Serpen and Yifeng X Electrical Engineering and Compter Science Department, The University of Toledo, Toledo, OH 4366 USA Abstract A new trainable and recrrent neral optimiation algorithm, which has potentially sperior capabilities compared to existing neral search algorithms to compte high qality soltions of static optimiation problems in a comptationally efficient manner, is stdied. Specifically, local stability analysis of the dynamics of a relaxation-based recrrent neral networ, the Simltaneos Recrrent Neral networ, for static optimiation problems is presented. The reslts of theoretical as well as its correlated simlation stdy lead to the conjectre that the Simltaneos Recrrent Neral networ dynamics appears to demonstrate desirable stability characteristics. Dynamics often converge to fixed points pon conclsion of a relaxation cycle, which facilitates adaptation of weights throgh one of many fixed-point training algorithms. The trainability of this neral algorithm reslts relatively high qality soltions to be compted for large-scale problem instances with comptational efficiency, particlarly when compared to soltions compted by the Hopfield networ and its derivative algorithms inclding those with stochastic search control mechanisms. I. INTRODUCTION Combinatorial optimiation is a field where crrently there are nmeros effective algorithms drawn from varios fields of science and engineering, among which most notable inclde heristic search within the domain of Artificial Intelligence, and Operations Research. Relatively large-scale problem instances were sccessflly addressed for near-optimal soltions by these non-neral algorithms. However, as the problem sie grows, the comptation time of a near-optimal soltion by almost all of these algorithms increases prohibitively. Artificial neral networs as a recent arrival in the field of combinatorial optimiation offers nparalleled scaleability for real-time comptation of soltions, and hence can deliver soltions in constant time for all practical prposes if implemented in hardware. This premise is inherently lacing in non-neral approaches de to facts that they lac any significant amont of parallelism and their implementation is realied on architectres that can offer large-grain parallelism at best. Hopfield family algorithms inclding the Boltmann Machine and the Mean-Field Annealing networ are significant neral algorithms for static optimiation [, 4, 7, 5]. The Simltaneos Recrrent Neral networ (SRNN was empirically demonstrated to offer sperior qalities to address large-scale instances of static optimiation problems compared to the Hopfield and its stochastic derivatives [5, 6,, 4]. Accordingly, a qestion that natrally follows is how SRNN is different than Hopfield family neral search algorithms to facilitate sch comptational capability. In a typical search scenario, a Hopfield networ and its derivatives wold be randomly initialied and relaxed towards a fixed point. If the fixed point the networ converges to is not a soltion, then another relaxation period with fresh and randomly initialied neron otpts is started discarding any experience that may have been associated with the prior relaxation. On the other hand, the SRNN is able to leverage the experience following an nsccessfl relaxation for the next one throgh its learning process. The SRNN individally adapts each and every weight in the networ sing a well-defined error fnction (in order to associate soltions of the problem with the fixed points of the networ dynamics in a spervised mode [4]. Extensive simlation stdies were performed to solve classical static optimiation problems inclding the Traveling Salesman problem (TSP and the Graph Path Search problem (GPSP with the Simltaneos Recrrent Neral networ [6,, 4]. Initially, the SRNN was configred as a relaxation networ and trained with the recrrent bacpropagation (RBP algorithm [2, 8, 2, 6]. Simlation reslts demonstrated that the SRNN was able to locate high qality soltions for instances of the TSP and GPSP p to 6 cities/vertices. The comptational cost of simlating SRNN/RBP algorithm on serial compting architectres was relatively significant yet considerably less than what it wold cost if sing a Hopfield family neral optimier [4]. Frther improvements for the SRNN by replacing the recrrent bacpropagation with the resilient propagation (RPROP algorithm broght abot very promising reslts inclding frther drastic improvements in the qality of soltions with no observable increase in the comptational cost [5]. In smmary, SRNN demonstrated two important advantages over what the Hopfield family algorithms cold offer: comptational ability to address large-scale static optimiation problems and, secondly, to locate high qality soltions for sch problems at a comparable, if not better, comptational cost.

2 The same simlation stdies also indicated throgh empirical means that the networ dynamics demonstrated globally asymptotically stable behavior in general. Specifically stable eqilibrim points were rotinely observed in the state space of the nonlinear networ dynamics. It wold be very comforting to identify a Liapnov fnction to theoretically validate the indications of the empirical stdy, which sggests stable nonlinear dynamics for the SRNN. However, this tas is not trivial and reqires mch trial-and-error. In the absence of a Liapnov fnction, sefl insight can be developed throgh a local stability analysis of the SRNN dynamics. A mathematical insight into stability properties of SRNN nonlinear dynamics is needed to assess its comptational promise and this paper presents sch a stdy. II. SIMULTANEOUS RECURRENT NEURAL NETWOR A mathematical characteriation of the comptation performed by a Simltaneos Recrrent Neral Networ (SRNN is given by [7] ˆ = f ( x, W, where x and W are the external inpts and weights, respectively and ẑ is the eqilibrim vale of following a relaxation cycle: ( n ˆ = lim, n which can be compted by the following iteration ( n+ ( n = f (, x,w, where f is a feedforward networ and n is the iteration index. Figre presents the high level topology of the SRNN. x FIGURE I SIMULTANEOUS RECURRENT NEURAL NETWOR f(,w The networ is provided with the external inpts and initial otpts, which are typically assmed randomly in the absence of a priori information. The otpt of previos iteration is fed bac to the networ along with the external inpts to compte the otpt of next iteration. The networ is allowed to iterate ntil it reaches a stable eqilibrim point assming at least one exists. External inpts are applied throghot the complete relaxation cycle. When a stable eqilibrim point is reached, the ( n+ otpts stop changing (i.e. the otpt vale of is (n almost eqal to or very close to. It is important to note that the feedbac from the otpt layer to inpt layer in SRNN is not delayed: the feedbac is, theoretically speaing, simltaneos. An SRNN exhibits complex temporal behavior: it follows a trajectory in the state space to relax to a fixed point. One relaxation of the networ consists of one or more iterations of otpt comptation and propagation along the feedforward and feedbac paths ntil the otpts converge to a stable eqilibrim vale. III. DYNAMIC SYSTEM ANALYSIS OF SRNN A two-layer topology for the SRNN was shown to be sfficient to represent soltions of problems in the domain of graph search problems inclding those in combinatorial optimiation i.e., the Traveling Salesman Problem (TSP. These two layers will be called the hidden layer and the otpt layer, where the hidden layer is a one-dimensional array of nodes and the otpt layer is a two-dimensional array of nodes[4]. This optimiation specific topology derives from the assmption that a mlti-layer continosdynamics perceptron networ with one hidden layer and no external inpts is instantiated for the feedforward mapping in Figre, which coincidentally reslts in the minimal strctre for SRNN. Symbols U and V will be employed to represent the and weight matrices associated with forward and bacward signal propagations, respectively. A. Local Stability Analysis for SRNN Dynamics Important nowledge abot the eqilibrim points for the dynamics of an SRNN can be dedced throgh a characteriation of the set of eqilibrim points and lineariation of the SRNN dynamics abot these eqilibrim points. More specifically, reqirements on the weight matrices to maintain stability of those eqilibrim points of interest for a given problem can be derived. This nderstanding is also liely to help infer conditions on the weight pdate rle nder a fixed-point training algorithm for SRNN dynamics. Let y = [ y y ] T and = [ ] T 2 y represent the node otpt vectors for the hidden and otpt layers, respectively, while the internal state of the otpt layer nerons is given by the vector s. Derivation of eqilibrim points for the SRNN dynamics will reqire an eqation of the form =, where is the vector of neron otpts in the otpt layer, which typically represents soltions for a given optimiation problem. This eqation reqires the 2

3 neron dynamics in the otpt layer to be represented by partial differential eqations. A minimm complexity model wold dictate these eqations to be of first-order and employ ideal integrators. A similar neron model was shown to address a set of important deficiencies of the neral networs composed of nerons employing lossy integrators []. However, comptational model for the nerons in the hidden layer can tilie continos perceptron eqations for the sae of simplicity and withot sacrificing generality. Hence, neron dynamics of SRNN are given by ds = f ( s with = dt for the nodes in the otpt layer, and j= y j = f ( net j with net j = v = for the nodes in the hidden layer, while noting that j y j j y j is the otpt of j-th neron in hidden layer, is the otpt of -th neron in otpt layer, is the node cont in hidden layer, is the nmber of nerons in the otpt layer, is the forward weight from j-th neron in hidden layer to - th neron in otpt layer, v is the bacward weight from j -th neron in otpt layer to j-th neron hidden layer, and f is continos, differentiable fnction typically a sigmoid with a steep slope (for combinatorial optimiation problems. For convenience, the activation fnction is instantiated as nipolar sigmoid with slope coefficient + λ R the same for both the hidden and otpt layers withot loss of generality. According to the chain rle, we have = s s for =, 2,,. Sbseqent to a series of algebraic maniplations, the matrix form of above eqation maps to [7]: = λ ( ( λ i = /( + e λ i = /( + e vi i v i i j ( Eqilibrim points of the networ dynamics are the soltions of the eqation determined by setting the vector of partial derivatives in Eqation to ero. Upon setting the leftmost matrix to a ero vector with appropriate dimensions, it is noted that if = or =, then ( = for all. Therefore, Eqation shows that the corners of the nit hypercbe are eqilibrim points in the state space of the SRNN dynamics. A second set of eqilibrim points can be obtained by solving the eqation that is obtained by setting the remaining part of above eqation to ero. Since the activation fnction of the nerons is assmed nipolar sigmoid, the otpt is continos. Soltions, which are also the eqilibrim points of SRNN dynamics, can be any points within the nit hypercbe in the otpt space. Assme that the high-gain mapping is employed for the sigmoid fnction, effectively forcing otpts of nerons to acqire limiting vales pon convergence. For a nipolar high-gain sigmoid fnction, the neron otpts are forced towards vales approaching. or.. In that case, the eqilibrim point set will inclde mostly (bt not necessarily only corners of the hypercbe. Next, the SRNN dynamics will be linearied at eqilibrim points, which inclde the corners of the hypercbe in the state space of the SRNN dynamics, to determine the stability properties of the same eqilibrim points. In order to derive the eigenvales at a corner of the hypercbe for the linearied dynamics, the SRNN dynamics are linearied arond a given hypercbe corner. be sch vertex with [ ] T Let = 2, = or, for =,2,,. Near the area of the corner of the hypercbe, consider a small deviation from by an amont sch that = +, for =,2,,. Sbseqently, it follows that eigenvales of the acobian matrix of the linearied dynamics are given by [7] j γ = λ( 2, (2, j= λ v jii i= + e where =,2,, and = or. The reqirement for linearied dynamics of the SRNN to be stable at a hypercbe corner is that all the eigenvales of the acobian matrix evalated at that eqilibrim point (hypercbe corner be negative:

4 γ < for =, 2,,., Referring to Eqation 2, we obtain T ( 2 y < for =,2,,, where = [ ] T and y is the otpt vector 2 of the hidden layer while noting that λ R +. In more specific terms, following conditions shold be satisfied in order to establish the stability of this eqilibrim point: and T j y < or < when j= λv jii + e i= i T j y > or > when j= + e λ v jii + v j i= i = =, where =, 2,,. The reslt is a set of simltaneos ineqalities, where the nnowns are weights for the forward and bacward signal propagations, and the soltion is non-trivial in order to determine bonds on these weights. However, the insight gained as a reslt of this analysis leads s to the following theorem. The theorem sggests the feasibility of establishing a given hypercbe corner in the state space of the SRNN dynamics as a stable eqilibrim point de to existence of real weight matrices U and V, which satisfy the set of ineqalities above for the given hypercbe corner. Theorem: Assme the SRNN dynamics is configred for static combinatorial optimiation: a two-layer topology, no external inpt, and high-gain sigmoid mapping for otpt layer nerons. Then, for a given hypercbe corner, there exist real weight matrices U and V to establish that hypercbe corner as a stable eqilibrim point in a local sense in the state space of the SRNN dynamics. Proof: A detailed derivation of the proof is presented in [8]. The proof for this theorem provides valable insight into the process of training the SRNN dynamics configred for static combinatorial optimiation sing a fixed-point learning algorithm. It demonstrates, with relative ease, the feasibility of associating the set of hypercbe corners that typically represent soltions of a combinatorial optimiation problem with stable eqilibrim points of the networ dynamics as the fixedpoint training algorithm searches throgh the weight space. The proof frther sggests hypercbe corners can easily be made stable eqilibrim points in the state space of SRNN dynamics throgh a relatively nconstrained selection of weights, which was also observed throghot a comprehensive set of simlation stdies [5, 6,, 4]. These reslts sggest that the SRNN as a combinatorial optimiation algorithm is robst with respect to changes in the adaptable parameters, namely weights, for its stability characteristics and therefore, facilitating reasonably nconstrained application of a learning algorithm inclding the recrrent bacpropagation as well as standard bacpropagation and its comptationally-efficient derivatives QicProp and resilient propagation [4, 3]. Some frther thoghts on global stability characteristics of the SRNN are detailed in the next section to demonstrate striing similarities among the SRNN, the Bidirectional Associative Memory (BAM, and the Adaptive Resonance Theory (ART networ cores. B. Global Stability and Liapnov Fnction for SRNN It is easy to show that Bidirectional Associative Memory (BAM is a special case of the SRNN, where the operational mode is associative memory and the feedforward networ for the SRNN is a single layer of perceptrons [9]. A BAM networ consists of two layers of nodes, labeled as a and b, and two weight matrices, M and M T, to facilitate the signal flow from nerons in one layer to another and vice versa. Nerons in a given layer do not interact among themselves, and employ bipolar threshold (or very high-gain sigmoid fnction for their otpts. A Liapnov fnction in the form of E T ( a, b = a b exists for the dynamics of a BAM networ and is valid for any real matrix M. T Consider setting U = V, where the forward weight matrix is eqal to the transpose of the bacward weight matrix, for the SRNN configred for static optimiation, which then leads to a typical BAM architectre. Frthermore, following simplifying assmptions are needed to transform an SRNN to a BAM networ: The forward mapping bloc in Figre is a twolayer feedforward perceptron networ, Perceptrons employ very high gain sigmoid mapping fnctions with bipolar otpts, and The external inpt for every neron/perceptron is ero. These observations sggest BAM, for which a Liapnov fnction exists to demonstrate global stability, is a special case of SRNN. Along similar lines, the Adaptive Resonance Theory & 2 (ART and ART2 neral networs are shown to possess a Liapnov fnction [3]. There exists a similarity between

5 the topological strctres and node dynamics of ART series networ cores and the SRNN configred for static optimiation. Given this similarity, well-established stability reslts for the ART series networs cold possibly pave the way to a more mathematically sond nderstanding and exploration of the stability properties of the SRNN dynamics. An empirical observation based on extensive simlation analysis performed on the TSP and graph search problems indicated that the SRNN is stable: it tends to converge to a fixed point starting with random initial vales for the weight matrices and neron otpt vales. Frthermore, the same simlation stdies also demonstrated that the application of the recrrent bacpropagation algorithm to train the weights does not appear to lead to instability in general. The SRNN algorithm is biologically inspired althogh it is most liely a very high-level and crde mathematical model at this stage [7]. However, this biological connection is an asset and motivating factor for the existence of a Liapnov fnction for the SRNN. These observations collectively form a sond platform to motivate the exploration of a Liapnov fnction for the SRNN. IV. SIMULATION STUDY The theorem presented earlier sggests that there exist real weight matrices U and V that establish a given hypercbe corner as a stable eqilibrim point in the state space of the nonlinear networ dynamics. The simlation stdy explores specific and niformly-sampled instances of U and V matrices in the agmented U-V space and assesses the set of stable eqilibrim points for a given instance of the U and V weight matrices with particlar emphasis on hypercbe corners as being sefl entities to represent soltions of combinatorial optimiation problems. If sfficient nmber of instances for the U and V matrices are generated throgh niform sampling for a given SRNN networ topology, it is reasonable to expect that the cmlative set of stable eqilibrim points encontered wold inclde the complete set of hypercbe corners. Ths, this finding wold validate the claim made by the theorem indicating that for a given hypercbe corner, U and V matrices exist to mae that hypercbe corner stable eqilibrim point. Consider a SRNN with the maximally redced topology for combinatorial optimiation: a single hidden layer with nodes, nodes in the otpt layer, and no external inpt. A high-gain nipolar sigmoid activation fnction with a steepness coefficient vale of is employed. The networ weights are randomly initialied to a vale between and with a niform probability distribtion. The initial otpt of the SRNN is set to be a hypercbe corner, which is also an eqilibrim point. The simlation is performed sing MATLAB version For the SRNN with small vales of, it is possible to test if all hypercbe corners are stable eqilibrim points for a given instance of U and V weight matrix pair. In the simlation stdy, a 2 4 SRNN, which is a networ topology with 2 nodes in the hidden layer and 4 nodes in the otpt layer, was stdied with respect to 75 different U and V weight matrix instances. The 2 4 SRNN has 2 4 possible hypercbe corners since there are 4 otpt nodes. Stability of all 6 hypercbe corners for the 75 instances of the U and V matrices was individally checed. Simlation reslts indicated that, ot of the 75 instances, in most cases at least one hypercbe corner was a stable eqilibrim point. All 6 hypercbe corners were observed to be stable eqilibrim points at least once for the 75 instances of weight matrices U and V. In most cases, the SRNN dynamics had a niqe stable eqilibrim point that was also a hypercbe corner. For some instances of U and V matrices, two hypercbe corners trned ot to be stable eqilibrim points. In a small portion of the 75 instances, limit cycles with a length of two appeared instead of a stable eqilibrim point. The otpt vector [ ] T was also a stable eqilibrim point for a small nmber of instances of U and V matrices. In smmary, 48 stable eqilibrim points that are also hypercbe corners, 2 limit cycles and 9 stable eqilibrim points that are non-hypercbe corners appeared throghot testing with 75 instances of U and V matrices in the simlation. Since three types of stable eqilibrim points (hypercbe corners, non-hypercbe corners, and limit cycles coexisted for some instances of U and V, the total nmber of stable eqilibrim points inclding limit cycles for this SRNN topology is larger than 75. Second simlation stdy was performed on a larger SRNN topology of 2 nodes. In the 2 SRNN case, since there are 2 hypercbe corners it is practically impossible to test the stability of every hypercbe corner eqilibrim point. Therefore, the simlation was condcted by randomly selecting hypercbe corner eqilibrim points among the set of 2 possibilities. Stability of these randomly selected eqilibrim points given a specific instance of weight matrices U and V was checed by forcing the SRNN throgh a relaxation while initialiing the otpt layer nerons to one of hypercbe corner eqilibrim points. The nmber of instances for the weight matrices U and V was, which translated into relaxations in total. In 997 ot of relaxations, a hypercbe corner eqilibrim point proved to be stable. The limit cycle, which had a cycle length of two, happened only once in the cases. The otpt vector

6 [ 5. 5 ] T..5 was also a stable eqilibrim point for two cases. Stdies on larger sie SRNNs involved the 5 and the 5 25 topology instances. A set of hypercbe corners was identified for each problem sie sing a niform random probability distribtion for stability testing given a particlar instance of the weight matrices. Nmber of instances of U and V matrices evalated for the 5 and the 5 25 instances of SRNN topology was. In total, cases were rn for each networ topology. In majority of cases, hypercbe corners appeared as stable eqilibrim points. In a fraction of cases, limit cycles were observed. More specifically, 99 hypercbe corners appeared as stable eqilibrim points while one limit cycle was observed for the 5 networ topology. The nmber of hypercbe corners as stable eqilibrim points was 97 with three limit cycles for the 5 25 networ topology. Simlation stdy indicated that any and all hypercbe corners cold potentially be stable eqilibrim points, which is controlled by the vales of weight matrices U and V. Two other types of eqilibrim points were also observed as attractors for the networ dynamics throghot simlations. These are the limit cycle and the eqilibrim point [ ] T for an SRNN with a - node otpt layer. The nmber of limit cycles and nonhypercbe eqilibrim points, which are attractors, is only a small fraction of the set of eqilibrim points, which can be established as stable points throgh the U and V matrices. V. CONCLUSIONS Theoretical local stability analysis of the Simltaneos Recrrent Neral Networ dynamics copled with correlated simlation stdy indicated that stable eqilibrim points, which inclde hypercbe corners as well as points on the edges, srfaces and interior of the hypercbe, exist in the state space of the neral networ dynamics. This facilitates the networ to be employed as a fixed-point attractor sitable for combinatorial optimiation. Frthermore, presence of sch stable eqilibrim points also facilitates employment of a fixedpoint training algorithm lie the recrrent bacpropagation to gide the networ towards the promising regions of the soltion space. The SRNN networ is comptationally mch more powerfl than the Hopfield networ and its derivatives to address challenging and large-scale combinatorial optimiation problems. Frther enhancement to the SRNN algorithm by incorporating a comptationally efficient stochastic search component is liely to be very valable to improve its ability to locate mch higher qality soltions for challenging problems. Acnowledgement - Fnding provided throgh United States National Science Fondation (NSF Grant (ECS for this research project is grateflly acnowledged. Opinions, views and conclsions expressed are athors only and do not reflect those of the NSF. BIBLIOGRAPHY [] Acley, D. H., Hinton, G. E., and Sejnowsi, T.., A Learning Algorithm for Boltman Machine, Cognitive Science, Vol. 9, pp , 985. [2] Almeida, L. B., A Learning Rle for Asynchronos Perceptrons with Feedbac in a Combinatorial Environment, Proceeding of IEEE -st International Conference on Neral Networs, San Diego, CA, ne 2-24, pp , 987. [3] Carpenter, G. and Grossberg, S., Adaptive Resonance Theory (ART, The Handboo of Brain Theory and Neral Networs, MIT Press, 995. [4] Cichoci, A. and Unbehaen, R., Neral Networ for Optimiation and Signal Processing, Wiley, 993. [5] Corra,, Training Simltaneos Recrrent Neral Networ with Non-Recrrent RPROP Training Algorithm, Internal Technical Report, The University of Toledo, Toledo, OH, 2. [6] Geib,., The Simltaneos Recrrent Neral Networ Applied to Large-Scale Traveling Salesman Problems, Internal Technical Report, Department of Electrical Engineering and Compter Science, The University of Toledo, Toledo, OH, 2. [7] Hopfield,.., and Tan, D. W., Neral Comptation of Decision in Optimiation Problems, Biological Cybernetics, Vol. 52, pp. 4-52, 985. [8] in, L., and Gpta, M. M., Stable Dynamic Bac-propagation Learning in Recrrent Neral Networs, IEEE Transactions on Neral Networs, Vol., No. 6, pp , 999. [9] oso, B., Bi-directional Associative Memories, IEEE Trans. on SMC, Vol. 8, pp. 49-6, 988. [] Li, -H., Michel, A. N., Porod, W., Analysis and Synthesis of A Class of Neral Networs: Linear Systems Operating on A Closed Hypercbe, IEEE Trans. On Circits and Systems, Vol. 36, No., pp , 989. [] Patwardhan, A., The Simltaneos Recrrent Neral Networ for Static Optimiation Problems, Master of Science in Engineering Science Thesis, The University of Toledo, 999. [2] Pineda, F.., Generaliation of Bac-Propagation to Recrrent Neral Networs, Physical Review Letters, Vol. 59, pp , 987. [3] Riedmiller, M. and Bran, H., A Direct Adaptive Method for Fast Bacpropagation Learning: The RPROP Algorithm, Proc. Of the IEEE International Conference on Neral Networs, Vol. 5, pp , 993. [4] Serpen, G., Patwardhan, A. and Geib,., The Simltaneos Recrrent Neral Networ Addressing the Scaling Problem in Static Optimiation, International ornal of Neral Systems, Vol., No. 5, pp , 2. [5] Serpen, G. and Livingston, D. L., Determination of Weights for Relaxation Recrrent Neral Networs, Nerocompting, Vol. 34, pp , 2. [6] Werbos, P.., Generaliation of Bacpropagation with Application to A Recrrent Gas Maret Model, Neral Networs, Vol., No. 4, pp , 988. [7] Werbos, P.., and Pang, X., Neral Networ Design for Fnction Approximation in Dynamic Programming, technical report available online at 8-/CSHCN_MS_98-.pdf, 998. [8] X, Y., A Dynamic System Analysis of Simltaneos Recrrent Neral Networ, Master of Science in Electrical Engineering Thesis, The University of Toledo, 2.

On the Computational Complexity and Effectiveness of N-hub Shortest-Path Routing

On the Computational Complexity and Effectiveness of N-hub Shortest-Path Routing 1 On the Comptational Complexity and Effectiveness of N-hb Shortest-Path Roting Reven Cohen Gabi Nakibli Dept. of Compter Sciences Technion Israel Abstract In this paper we stdy the comptational complexity

More information

TRAINING SIMULTANEOUS RECURRENT NEURAL NETWORK WITH RESILIENT PROPAGATION FOR STATIC OPTIMIZATION

TRAINING SIMULTANEOUS RECURRENT NEURAL NETWORK WITH RESILIENT PROPAGATION FOR STATIC OPTIMIZATION TRAIIG SIMULTAEOUS RECURRET EURAL ETWORK WITH RESILIET PROPAGATIO FOR STATIC OPTIMIZATIO Gursel Serpen * and Joel Corra Electrical Engineering and Computer Science, The University of Toledo, Toledo, OH

More information

Master for Co-Simulation Using FMI

Master for Co-Simulation Using FMI Master for Co-Simlation Using FMI Jens Bastian Christoph Claß Ssann Wolf Peter Schneider Franhofer Institte for Integrated Circits IIS / Design Atomation Division EAS Zenerstraße 38, 69 Dresden, Germany

More information

PARAMETER OPTIMIZATION FOR TAKAGI-SUGENO FUZZY MODELS LESSONS LEARNT

PARAMETER OPTIMIZATION FOR TAKAGI-SUGENO FUZZY MODELS LESSONS LEARNT PAAMETE OPTIMIZATION FO TAKAGI-SUGENO FUZZY MODELS LESSONS LEANT Manfred Männle Inst. for Compter Design and Falt Tolerance Univ. of Karlsrhe, 768 Karlsrhe, Germany maennle@compter.org Brokat Technologies

More information

An Adaptive Strategy for Maximizing Throughput in MAC layer Wireless Multicast

An Adaptive Strategy for Maximizing Throughput in MAC layer Wireless Multicast University of Pennsylvania ScholarlyCommons Departmental Papers (ESE) Department of Electrical & Systems Engineering May 24 An Adaptive Strategy for Maximizing Throghpt in MAC layer Wireless Mlticast Prasanna

More information

Networks An introduction to microcomputer networking concepts

Networks An introduction to microcomputer networking concepts Behavior Research Methods& Instrmentation 1978, Vol 10 (4),522-526 Networks An introdction to microcompter networking concepts RALPH WALLACE and RICHARD N. JOHNSON GA TX, Chicago, Illinois60648 and JAMES

More information

Evaluating Influence Diagrams

Evaluating Influence Diagrams Evalating Inflence Diagrams Where we ve been and where we re going Mark Crowley Department of Compter Science University of British Colmbia crowley@cs.bc.ca Agst 31, 2004 Abstract In this paper we will

More information

Nash Convergence of Gradient Dynamics in General-Sum Games. Michael Kearns.

Nash Convergence of Gradient Dynamics in General-Sum Games. Michael Kearns. Convergence of Gradient Dynamics in General-Sm Games Satinder Singh AT&T Labs Florham Park, NJ 7932 bavejaresearch.att.com Michael Kearns AT&T Labs Florham Park, NJ 7932 mkearnsresearch.att.com Yishay

More information

Multiple-Choice Test Chapter Golden Section Search Method Optimization COMPLETE SOLUTION SET

Multiple-Choice Test Chapter Golden Section Search Method Optimization COMPLETE SOLUTION SET Mltiple-Choice Test Chapter 09.0 Golden Section Search Method Optimization COMPLETE SOLUTION SET. Which o the ollowing statements is incorrect regarding the Eqal Interval Search and Golden Section Search

More information

Optimal Sampling in Compressed Sensing

Optimal Sampling in Compressed Sensing Optimal Sampling in Compressed Sensing Joyita Dtta Introdction Compressed sensing allows s to recover objects reasonably well from highly ndersampled data, in spite of violating the Nyqist criterion. In

More information

Resolving Linkage Anomalies in Extracted Software System Models

Resolving Linkage Anomalies in Extracted Software System Models Resolving Linkage Anomalies in Extracted Software System Models Jingwei W and Richard C. Holt School of Compter Science University of Waterloo Waterloo, Canada j25w, holt @plg.waterloo.ca Abstract Program

More information

Pavlin and Daniel D. Corkill. Department of Computer and Information Science University of Massachusetts Amherst, Massachusetts 01003

Pavlin and Daniel D. Corkill. Department of Computer and Information Science University of Massachusetts Amherst, Massachusetts 01003 From: AAAI-84 Proceedings. Copyright 1984, AAAI (www.aaai.org). All rights reserved. SELECTIVE ABSTRACTION OF AI SYSTEM ACTIVITY Jasmina Pavlin and Daniel D. Corkill Department of Compter and Information

More information

IMPLEMENTATION OF OBJECT ORIENTED APPROACH TO MODIFIED ANT ALGORITHM FOR TASK SCHEDULING IN GRID COMPUTING

IMPLEMENTATION OF OBJECT ORIENTED APPROACH TO MODIFIED ANT ALGORITHM FOR TASK SCHEDULING IN GRID COMPUTING International Jornal of Modern Engineering Research (IJMER) www.imer.com Vol.1, Isse1, pp-134-139 ISSN: 2249-6645 IMPLEMENTATION OF OBJECT ORIENTED APPROACH TO MODIFIED ANT ALGORITHM FOR TASK SCHEDULING

More information

IDENTIFICATION OF THE AEROELASTIC MODEL OF A LARGE TRANSPORT CIVIL AIRCRAFT FOR CONTROL LAW DESIGN AND VALIDATION

IDENTIFICATION OF THE AEROELASTIC MODEL OF A LARGE TRANSPORT CIVIL AIRCRAFT FOR CONTROL LAW DESIGN AND VALIDATION ICAS 2 CONGRESS IDENTIFICATION OF THE AEROELASTIC MODEL OF A LARGE TRANSPORT CIVIL AIRCRAFT FOR CONTROL LAW DESIGN AND VALIDATION Christophe Le Garrec, Marc Hmbert, Michel Lacabanne Aérospatiale Matra

More information

The Impact of Avatar Mobility on Distributed Server Assignment for Delivering Mobile Immersive Communication Environment

The Impact of Avatar Mobility on Distributed Server Assignment for Delivering Mobile Immersive Communication Environment This fll text paper was peer reviewed at the direction of IEEE Commnications Society sbject matter experts for pblication in the ICC 27 proceedings. The Impact of Avatar Mobility on Distribted Server Assignment

More information

FINITE ELEMENT APPROXIMATION OF CONVECTION DIFFUSION PROBLEMS USING GRADED MESHES

FINITE ELEMENT APPROXIMATION OF CONVECTION DIFFUSION PROBLEMS USING GRADED MESHES FINITE ELEMENT APPROXIMATION OF CONVECTION DIFFUSION PROBLEMS USING GRADED MESHES RICARDO G. DURÁN AND ARIEL L. LOMBARDI Abstract. We consider the nmerical approximation of a model convection-diffsion

More information

REDUCED-REFERENCE ASSESSMENT OF PERCEIVED QUALITY BY EXPLOITING COLOR INFORMATION

REDUCED-REFERENCE ASSESSMENT OF PERCEIVED QUALITY BY EXPLOITING COLOR INFORMATION REDUCED-REFERENCE ASSESSMENT OF PERCEIVED QUALITY BY EXPLOITING COLOR INFORMATION Jdith Redi, Paolo Gastaldo, Rodolfo Znino, and Ingrid Heyndericx (+) University of Genoa, DIBE, Via Opera Pia a - 645 Genova

More information

COMPOSITION OF STABLE SET POLYHEDRA

COMPOSITION OF STABLE SET POLYHEDRA COMPOSITION OF STABLE SET POLYHEDRA Benjamin McClosky and Illya V. Hicks Department of Comptational and Applied Mathematics Rice University November 30, 2007 Abstract Barahona and Mahjob fond a defining

More information

Tu P7 15 First-arrival Traveltime Tomography with Modified Total Variation Regularization

Tu P7 15 First-arrival Traveltime Tomography with Modified Total Variation Regularization T P7 15 First-arrival Traveltime Tomography with Modified Total Variation Reglarization W. Jiang* (University of Science and Technology of China) & J. Zhang (University of Science and Technology of China)

More information

A sufficient condition for spiral cone beam long object imaging via backprojection

A sufficient condition for spiral cone beam long object imaging via backprojection A sfficient condition for spiral cone beam long object imaging via backprojection K. C. Tam Siemens Corporate Research, Inc., Princeton, NJ, USA Abstract The response of a point object in cone beam spiral

More information

REPLICATION IN BANDWIDTH-SYMMETRIC BITTORRENT NETWORKS. M. Meulpolder, D.H.J. Epema, H.J. Sips

REPLICATION IN BANDWIDTH-SYMMETRIC BITTORRENT NETWORKS. M. Meulpolder, D.H.J. Epema, H.J. Sips REPLICATION IN BANDWIDTH-SYMMETRIC BITTORRENT NETWORKS M. Melpolder, D.H.J. Epema, H.J. Sips Parallel and Distribted Systems Grop Department of Compter Science, Delft University of Technology, the Netherlands

More information

POWER-OF-2 BOUNDARIES

POWER-OF-2 BOUNDARIES Warren.3.fm Page 5 Monday, Jne 17, 5:6 PM CHAPTER 3 POWER-OF- BOUNDARIES 3 1 Ronding Up/Down to a Mltiple of a Known Power of Ronding an nsigned integer down to, for eample, the net smaller mltiple of

More information

Discretized Approximations for POMDP with Average Cost

Discretized Approximations for POMDP with Average Cost Discretized Approximations for POMDP with Average Cost Hizhen Y Lab for Information and Decisions EECS Dept., MIT Cambridge, MA 0239 Dimitri P. Bertsekas Lab for Information and Decisions EECS Dept., MIT

More information

StaCo: Stackelberg-based Coverage Approach in Robotic Swarms

StaCo: Stackelberg-based Coverage Approach in Robotic Swarms Maastricht University Department of Knowledge Engineering Technical Report No.:... : Stackelberg-based Coverage Approach in Robotic Swarms Kateřina Staňková, Bijan Ranjbar-Sahraei, Gerhard Weiss, Karl

More information

The Disciplined Flood Protocol in Sensor Networks

The Disciplined Flood Protocol in Sensor Networks The Disciplined Flood Protocol in Sensor Networks Yong-ri Choi and Mohamed G. Goda Department of Compter Sciences The University of Texas at Astin, U.S.A. fyrchoi, godag@cs.texas.ed Hssein M. Abdel-Wahab

More information

Maximum Weight Independent Sets in an Infinite Plane

Maximum Weight Independent Sets in an Infinite Plane Maximm Weight Independent Sets in an Infinite Plane Jarno Nosiainen, Jorma Virtamo, Pasi Lassila jarno.nosiainen@tkk.fi, jorma.virtamo@tkk.fi, pasi.lassila@tkk.fi Department of Commnications and Networking

More information

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 5, MAY On the Analysis of the Bluetooth Time Division Duplex Mechanism

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 5, MAY On the Analysis of the Bluetooth Time Division Duplex Mechanism IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 6, NO. 5, MAY 2007 1 On the Analysis of the Bletooth Time Division Dplex Mechanism Gil Zssman Member, IEEE, Adrian Segall Fellow, IEEE, and Uri Yechiali

More information

Fault Tolerance in Hypercubes

Fault Tolerance in Hypercubes Falt Tolerance in Hypercbes Shobana Balakrishnan, Füsn Özgüner, and Baback A. Izadi Department of Electrical Engineering, The Ohio State University, Colmbs, OH 40, USA Abstract: This paper describes different

More information

Switched state-feedback controllers with multi-estimators for MIMO systems

Switched state-feedback controllers with multi-estimators for MIMO systems Proceedings of the th WEA Int Conf on COMPUTATIONAL INTELLIGENCE MAN-MACHINE YTEM AND CYBERNETIC Venice Ital November - 6 89 witched state-feedback controllers with mlti-estimators for MIMO sstems LIBOR

More information

Discrete Cost Multicommodity Network Optimization Problems and Exact Solution Methods

Discrete Cost Multicommodity Network Optimization Problems and Exact Solution Methods Annals of Operations Research 106, 19 46, 2001 2002 Klwer Academic Pblishers. Manfactred in The Netherlands. Discrete Cost Mlticommodity Network Optimization Problems and Exact Soltion Methods MICHEL MINOUX

More information

Hardware-Accelerated Free-Form Deformation

Hardware-Accelerated Free-Form Deformation Hardware-Accelerated Free-Form Deformation Clint Cha and Ulrich Nemann Compter Science Department Integrated Media Systems Center University of Sothern California Abstract Hardware-acceleration for geometric

More information

Minimal Edge Addition for Network Controllability

Minimal Edge Addition for Network Controllability This article has been accepted for pblication in a ftre isse of this jornal, bt has not been flly edited. Content may change prior to final pblication. Citation information: DOI 10.1109/TCNS.2018.2814841,

More information

Appearance Based Tracking with Background Subtraction

Appearance Based Tracking with Background Subtraction The 8th International Conference on Compter Science & Edcation (ICCSE 213) April 26-28, 213. Colombo, Sri Lanka SD1.4 Appearance Based Tracking with Backgrond Sbtraction Dileepa Joseph Jayamanne Electronic

More information

The final datapath. M u x. Add. 4 Add. Shift left 2. PCSrc. RegWrite. MemToR. MemWrite. Read data 1 I [25-21] Instruction. Read. register 1 Read.

The final datapath. M u x. Add. 4 Add. Shift left 2. PCSrc. RegWrite. MemToR. MemWrite. Read data 1 I [25-21] Instruction. Read. register 1 Read. The final path PC 4 Add Reg Shift left 2 Add PCSrc Instrction [3-] Instrction I [25-2] I [2-6] I [5 - ] register register 2 register 2 Registers ALU Zero Reslt ALUOp em Data emtor RegDst ALUSrc em I [5

More information

Computer-Aided Mechanical Design Using Configuration Spaces

Computer-Aided Mechanical Design Using Configuration Spaces Compter-Aided Mechanical Design Using Configration Spaces Leo Joskowicz Institte of Compter Science The Hebrew University Jersalem 91904, Israel E-mail: josko@cs.hji.ac.il Elisha Sacks (corresponding athor)

More information

A GENERIC MODEL OF A BASE-ISOLATED BUILDING

A GENERIC MODEL OF A BASE-ISOLATED BUILDING Chapter 5 A GENERIC MODEL OF A BASE-ISOLATED BUILDING This chapter draws together the work o Chapters 3 and 4 and describes the assembly o a generic model o a base-isolated bilding. The irst section describes

More information

An Optimization of Granular Network by Evolutionary Methods

An Optimization of Granular Network by Evolutionary Methods An Optimization of Granlar Networ by Evoltionary Methods YUN-HEE HAN, KEUN-CHANG KWAK* Dept. of Control, Instrmentation, and Robot Engineering Chosn University 375 Seos-dong, Dong-g, Gwangj, 50-759 Soth

More information

Topic Continuity for Web Document Categorization and Ranking

Topic Continuity for Web Document Categorization and Ranking Topic Continity for Web ocment Categorization and Ranking B. L. Narayan, C. A. Mrthy and Sankar. Pal Machine Intelligence Unit, Indian Statistical Institte, 03, B. T. Road, olkata - 70008, India. E-mail:

More information

Method to build an initial adaptive Neuro-Fuzzy controller for joints control of a legged robot

Method to build an initial adaptive Neuro-Fuzzy controller for joints control of a legged robot Method to bild an initial adaptive Nero-Fzzy controller for joints control of a legged robot J-C Habmremyi, P. ool and Y. Badoin Royal Military Academy-Free University of Brssels 08 Hobbema str, box:mrm,

More information

Prof. Kozyrakis. 1. (10 points) Consider the following fragment of Java code:

Prof. Kozyrakis. 1. (10 points) Consider the following fragment of Java code: EE8 Winter 25 Homework #2 Soltions De Thrsday, Feb 2, 5 P. ( points) Consider the following fragment of Java code: for (i=; i

More information

Image Denoising Algorithms

Image Denoising Algorithms Image Denoising Algorithms Xiang Hao School of Compting, University of Utah, USA, hao@cs.tah.ed Abstract. This is a report of an assignment of the class Mathematics of Imaging. In this assignment, we first

More information

Augmenting the edge connectivity of planar straight line graphs to three

Augmenting the edge connectivity of planar straight line graphs to three Agmenting the edge connectivity of planar straight line graphs to three Marwan Al-Jbeh Mashhood Ishaqe Kristóf Rédei Diane L. Sovaine Csaba D. Tóth Pavel Valtr Abstract We characterize the planar straight

More information

Picking and Curves Week 6

Picking and Curves Week 6 CS 48/68 INTERACTIVE COMPUTER GRAPHICS Picking and Crves Week 6 David Breen Department of Compter Science Drexel University Based on material from Ed Angel, University of New Mexico Objectives Picking

More information

CS 4204 Computer Graphics

CS 4204 Computer Graphics CS 424 Compter Graphics Crves and Srfaces Yong Cao Virginia Tech Reference: Ed Angle, Interactive Compter Graphics, University of New Mexico, class notes Crve and Srface Modeling Objectives Introdce types

More information

A choice relation framework for supporting category-partition test case generation

A choice relation framework for supporting category-partition test case generation Title A choice relation framework for spporting category-partition test case generation Athor(s) Chen, TY; Poon, PL; Tse, TH Citation Ieee Transactions On Software Engineering, 2003, v. 29 n. 7, p. 577-593

More information

Semantic segmentation is the task of labeling every pixel in

Semantic segmentation is the task of labeling every pixel in Deep Learning for Visal Understanding: Part Anrag Arnab, Shai Zheng, Sadeep Jayasmana, Bernardino Romera-Paredes, Måns Larsson, Alexander Kirillov, Bogdan Savchynskyy, Carsten Rother, Fredrik Kahl, and

More information

Factors Influencing Performance of Firefly and Particle Swarm Optimization Algorithms

Factors Influencing Performance of Firefly and Particle Swarm Optimization Algorithms Factors Inflencing Performance of Firefly and Particle Swarm Optimization Algorithms Damanjeet Kar, UIET, Panjab University, Chandigarh Abstract In this paper, two natre inspired meta heristic approaches

More information

Bias of Higher Order Predictive Interpolation for Sub-pixel Registration

Bias of Higher Order Predictive Interpolation for Sub-pixel Registration Bias of Higher Order Predictive Interpolation for Sb-pixel Registration Donald G Bailey Institte of Information Sciences and Technology Massey University Palmerston North, New Zealand D.G.Bailey@massey.ac.nz

More information

Real-time mean-shift based tracker for thermal vision systems

Real-time mean-shift based tracker for thermal vision systems 9 th International Conference on Qantitative InfraRed Thermography Jly -5, 008, Krakow - Poland Real-time mean-shift based tracker for thermal vision systems G. Bieszczad* T. Sosnowski** * Military University

More information

Improved Estimation of Transmission Distortion for Error-resilient Video Coding

Improved Estimation of Transmission Distortion for Error-resilient Video Coding 1 Improved Estimation of Transmission Distortion for Error-resilient Video Coding Zhifeng Chen 1, Peshala Pahalawatta 2, Alexis Michael Torapis 2, and Dapeng W 1,* 1 Department of Electrical and Compter

More information

The effectiveness of PIES comparing to FEM and BEM for 3D elasticity problems

The effectiveness of PIES comparing to FEM and BEM for 3D elasticity problems The effectiveness of PIES comparing to FEM and BEM for 3D elasticit problems A. Boltc, E. Zienik, K. Sersen Faclt of Mathematics and Compter Science, Universit of Bialstok, Sosnowa 64, 15-887 Bialstok,

More information

Tdb: A Source-level Debugger for Dynamically Translated Programs

Tdb: A Source-level Debugger for Dynamically Translated Programs Tdb: A Sorce-level Debgger for Dynamically Translated Programs Naveen Kmar, Brce R. Childers, and Mary Lo Soffa Department of Compter Science University of Pittsbrgh Pittsbrgh, Pennsylvania 15260 {naveen,

More information

An Introduction to GPU Computing. Aaron Coutino MFCF

An Introduction to GPU Computing. Aaron Coutino MFCF An Introdction to GPU Compting Aaron Cotino acotino@waterloo.ca MFCF What is a GPU? A GPU (Graphical Processing Unit) is a special type of processor that was designed to render and maniplate textres. They

More information

Fast Obstacle Detection using Flow/Depth Constraint

Fast Obstacle Detection using Flow/Depth Constraint Fast Obstacle etection sing Flow/epth Constraint S. Heinrich aimlerchrylser AG P.O.Box 2360, -89013 Ulm, Germany Stefan.Heinrich@aimlerChrysler.com Abstract The early recognition of potentially harmfl

More information

Broadcasting XORs: On the Application of Network Coding in Access Point-to-Multipoint Networks

Broadcasting XORs: On the Application of Network Coding in Access Point-to-Multipoint Networks Broadcasting XORs: On the Application of Network Coding in Access Point-to-Mltipoint Networks The MIT Faclty has made this article openly available Please share how this access benefits yo Yor story matters

More information

Blended Deformable Models

Blended Deformable Models Blended Deformable Models (In IEEE Trans. Pattern Analysis and Machine Intelligence, April 996, 8:4, pp. 443-448) Doglas DeCarlo and Dimitri Metaxas Department of Compter & Information Science University

More information

Today s Lecture. Software Architecture. Lecture 27: Introduction to Software Architecture. Introduction and Background of

Today s Lecture. Software Architecture. Lecture 27: Introduction to Software Architecture. Introduction and Background of Today s Lectre Lectre 27: Introdction to Software Architectre Kenneth M. Anderson Fondations of Software Engineering CSCI 5828 - Spring Semester, 1999 Introdction and Backgrond of Software Architectre

More information

Requirements Engineering. Objectives. System requirements. Types of requirements. FAQS about requirements. Requirements problems

Requirements Engineering. Objectives. System requirements. Types of requirements. FAQS about requirements. Requirements problems Reqirements Engineering Objectives An introdction to reqirements Gerald Kotonya and Ian Sommerville To introdce the notion of system reqirements and the reqirements process. To explain how reqirements

More information

arxiv: v1 [cs.cr] 3 Jun 2017

arxiv: v1 [cs.cr] 3 Jun 2017 Spectrm-based deep neral networks for frad detection arxiv:1706.00891v1 [cs.cr] 3 Jn 2017 ABSTRACT Shhan Yan Tongji University 4e66@tongji.ed.cn Jn Li University of Oregon lijn@cs.oregon.ed In this paper,

More information

This chapter is based on the following sources, which are all recommended reading:

This chapter is based on the following sources, which are all recommended reading: Bioinformatics I, WS 09-10, D. Hson, December 7, 2009 105 6 Fast String Matching This chapter is based on the following sorces, which are all recommended reading: 1. An earlier version of this chapter

More information

ABSOLUTE DEFORMATION PROFILE MEASUREMENT IN TUNNELS USING RELATIVE CONVERGENCE MEASUREMENTS

ABSOLUTE DEFORMATION PROFILE MEASUREMENT IN TUNNELS USING RELATIVE CONVERGENCE MEASUREMENTS Proceedings th FIG Symposim on Deformation Measrements Santorini Greece 00. ABSOUTE DEFORMATION PROFIE MEASUREMENT IN TUNNES USING REATIVE CONVERGENCE MEASUREMENTS Mahdi Moosai and Saeid Khazaei Mining

More information

THE Unit Commitment problem (UCP) is the problem of

THE Unit Commitment problem (UCP) is the problem of IEEE TRANS IN POWER SYSTEMS 1 A new MILP-based approach for Unit Commitment in power prodction planning. Ana Viana and João Pedro Pedroso Abstract This paper presents a novel, iterative optimisation algorithm

More information

Pipelined van Emde Boas Tree: Algorithms, Analysis, and Applications

Pipelined van Emde Boas Tree: Algorithms, Analysis, and Applications This fll text paper was peer reviewed at the direction of IEEE Commnications Society sbject matter experts for pblication in the IEEE INFOCOM 007 proceedings Pipelined van Emde Boas Tree: Algorithms, Analysis,

More information

Date: December 5, 1999 Dist'n: T1E1.4

Date: December 5, 1999 Dist'n: T1E1.4 12/4/99 1 T1E14/99-559 Project: T1E14: VDSL Title: Vectored VDSL (99-559) Contact: J Cioffi, G Ginis, W Y Dept of EE, Stanford U, Stanford, CA 945 Cioffi@stanforded, 1-65-723-215, F: 1-65-724-3652 Date:

More information

Millimeter-Wave Multi-Hop Wireless Backhauling for 5G Cellular Networks

Millimeter-Wave Multi-Hop Wireless Backhauling for 5G Cellular Networks 2017 IEEE 85th Vehiclar Technology Conference (VTC-Spring) Millimeter-Wave Mlti-Hop Wireless Backhaling for 5G Celllar Networks B. P. S. Sahoo, Chn-Han Yao, and Hng-Y Wei Gradate Institte of Electrical

More information

Continuity Smooth Path Planning Using Cubic Polynomial Interpolation with Membership Function

Continuity Smooth Path Planning Using Cubic Polynomial Interpolation with Membership Function J Electr Eng Technol Vol., No.?: 74-?, 5 http://dx.doi.org/.537/jeet.5..?.74 ISSN(Print) 975- ISSN(Online) 93-743 Continity Smooth Path Planning Using Cbic Polomial Interpolation with Membership Fnction

More information

METAMODEL FOR SOFTWARE SOLUTIONS IN COMPUTED TOMOGRAPHY

METAMODEL FOR SOFTWARE SOLUTIONS IN COMPUTED TOMOGRAPHY VOL. 10, NO 22, DECEBER, 2015 ISSN 1819-6608 ETAODEL FOR SOFTWARE SOLUTIONS IN COPUTED TOOGRAPHY Vitaliy ezhyev Faclty of Compter Systems and Software Engineering, Universiti alaysia Pahang, Gambang, alaysia

More information

Subgraph Matching with Set Similarity in a Large Graph Database

Subgraph Matching with Set Similarity in a Large Graph Database 1 Sbgraph Matching with Set Similarity in a Large Graph Database Liang Hong, Lei Zo, Xiang Lian, Philip S. Y Abstract In real-world graphs sch as social networks, Semantic Web and biological networks,

More information

Image Compression Compression Fundamentals

Image Compression Compression Fundamentals Compression Fndamentals Data compression refers to the process of redcing the amont of data reqired to represent given qantity of information. Note that data and information are not the same. Data refers

More information

Curves and Surfaces. CS 537 Interactive Computer Graphics Prof. David E. Breen Department of Computer Science

Curves and Surfaces. CS 537 Interactive Computer Graphics Prof. David E. Breen Department of Computer Science Crves and Srfaces CS 57 Interactive Compter Graphics Prof. David E. Breen Department of Compter Science E. Angel and D. Shreiner: Interactive Compter Graphics 6E Addison-Wesley 22 Objectives Introdce types

More information

History Slicing: Assisting Code-Evolution Tasks

History Slicing: Assisting Code-Evolution Tasks History Slicing: Assisting Code-Evoltion Tasks Francisco Servant Department of Informatics University of California, Irvine Irvine, CA, U.S.A. 92697-3440 fservant@ics.ci.ed James A. Jones Department of

More information

Neighbourhood Searches for the Bounded Diameter Minimum Spanning Tree Problem Embedded in a VNS, EA, and ACO

Neighbourhood Searches for the Bounded Diameter Minimum Spanning Tree Problem Embedded in a VNS, EA, and ACO Neighborhood Searches for the Bonded Diameter Minimm Spanning Tree Problem Embedded in a VNS, EA, and ACO Martin Grber Institte for Compter Graphics and Algorithms, Vienna University of Technology, Favoritenstraße

More information

Adaptive Influence Maximization in Microblog under the Competitive Independent Cascade Model

Adaptive Influence Maximization in Microblog under the Competitive Independent Cascade Model International Jornal of Knowledge Engineering, Vol. 1, No. 2, September 215 Adaptie Inflence Maximization in Microblog nder the Competitie Independent Cascade Model Zheng Ding, Kai Ni, and Zhiqiang He

More information

Towards Understanding Bilevel Multi-objective Optimization with Deterministic Lower Level Decisions

Towards Understanding Bilevel Multi-objective Optimization with Deterministic Lower Level Decisions Towards Understanding Bilevel Mlti-objective Optimization with Deterministic Lower Level Decisions Ankr Sinha Ankr.Sinha@aalto.fi Department of Information and Service Economy, Aalto University School

More information

IoT-Cloud Service Optimization in Next Generation Smart Environments

IoT-Cloud Service Optimization in Next Generation Smart Environments 1 IoT-Clod Service Optimization in Next Generation Smart Environments Marc Barcelo, Alejandro Correa, Jaime Llorca, Antonia M. Tlino, Jose Lopez Vicario, Antoni Morell Universidad Atonoma de Barcelona,

More information

New Architectures for Hierarchical Predictive Control

New Architectures for Hierarchical Predictive Control Preprint, 11th IFAC Symposim on Dynamics and Control of Process Systems, inclding Biosystems Jne 6-8, 216. NTNU, Trondheim, Norway New Architectres for Hierarchical Predictive Control Victor M. Zavala

More information

Improving Network Connectivity Using Trusted Nodes and Edges

Improving Network Connectivity Using Trusted Nodes and Edges Improing Network Connectiity Using Trsted Nodes and Edges Waseem Abbas, Aron Laszka, Yegeniy Vorobeychik, and Xenofon Kotsokos Abstract Network connectiity is a primary attribte and a characteristic phenomenon

More information

A Wireless MAC Protocol comparison.

A Wireless MAC Protocol comparison. Doc: IEEE P82.1192S1 IEEE 82.11 Wireless Access Method and Physical Layer Specifications Title: Presented by: A Wireless MAC Protocol comparison. Wim Diepstraten NCR SEUtrecht NCRAT&T Network Prodct Grop

More information

SZ-1.4: Significantly Improving Lossy Compression for Scientific Data Sets Based on Multidimensional Prediction and Error- Controlled Quantization

SZ-1.4: Significantly Improving Lossy Compression for Scientific Data Sets Based on Multidimensional Prediction and Error- Controlled Quantization SZ-1.4: Significantly Improving Lossy Compression for Scientific Data Sets Based on Mltidimensional Prediction and Error- Controlled Qantization Dingwen Tao (University of California, Riverside) Sheng

More information

Advance Fuzzy Programming Approach to Bi-Level Quadratic Fractional Programming Problems by Genetic Algorithm

Advance Fuzzy Programming Approach to Bi-Level Quadratic Fractional Programming Problems by Genetic Algorithm American International Jornal of Research in Science, Technology, Engineering & Mathematics Available online at http://www.iasir.net ISSN (Print): 38-349, ISSN (Online): 38-3580, ISSN (CD-ROM): 38-369

More information

Abstract 1 Introduction

Abstract 1 Introduction Combining Relevance Information in a Synchronos Collaborative Information Retrieval Environment Colm Foley, Alan F. Smeaton and Gareth J. F. Jones Centre for Digital Video Processing and Adaptive Information

More information

New-Sum: A Novel Online ABFT Scheme For General Iterative Methods

New-Sum: A Novel Online ABFT Scheme For General Iterative Methods New-Sm: A Novel Online ABFT Scheme For General Iterative Methods Dingwen Tao (University of California, Riverside) Shaiwen Leon Song (Pacific Northwest National Laboratory) Sriram Krishnamoorthy (Pacific

More information

Congestion-adaptive Data Collection with Accuracy Guarantee in Cyber-Physical Systems

Congestion-adaptive Data Collection with Accuracy Guarantee in Cyber-Physical Systems Congestion-adaptive Data Collection with Accracy Garantee in Cyber-Physical Systems Nematollah Iri, Lei Y, Haiying Shen, Gregori Calfield Department of Electrical and Compter Engineering, Clemson University,

More information

h-vectors of PS ear-decomposable graphs

h-vectors of PS ear-decomposable graphs h-vectors of PS ear-decomposable graphs Nima Imani 2, Lee Johnson 1, Mckenzie Keeling-Garcia 1, Steven Klee 1 and Casey Pinckney 1 1 Seattle University Department of Mathematics, 901 12th Avene, Seattle,

More information

Computer User s Guide 4.0

Computer User s Guide 4.0 Compter User s Gide 4.0 2001 Glenn A. Miller, All rights reserved 2 The SASSI Compter User s Gide 4.0 Table of Contents Chapter 1 Introdction...3 Chapter 2 Installation and Start Up...5 System Reqirements

More information

Scientific Computing on Bulk Synchronous Parallel Architectures

Scientific Computing on Bulk Synchronous Parallel Architectures Scientific Compting on Blk Synchronos Parallel Architectres R. H. Bisseling Department of Mathematics, Utrecht University and W. F. McColl y Programming Research Grop, Oxford University April 27, 1994

More information

Alliances and Bisection Width for Planar Graphs

Alliances and Bisection Width for Planar Graphs Alliances and Bisection Width for Planar Graphs Martin Olsen 1 and Morten Revsbæk 1 AU Herning Aarhs University, Denmark. martino@hih.a.dk MADAGO, Department of Compter Science Aarhs University, Denmark.

More information

IP Multicast Fault Recovery in PIM over OSPF

IP Multicast Fault Recovery in PIM over OSPF 1 IP Mlticast Falt Recovery in PIM over OSPF Abstract Relatively little attention has been given to nderstanding the falt recovery characteristics and performance tning of native IP mlticast networks.

More information

AUTOMATIC REGISTRATION FOR REPEAT-TRACK INSAR DATA PROCESSING

AUTOMATIC REGISTRATION FOR REPEAT-TRACK INSAR DATA PROCESSING AUTOMATIC REGISTRATION FOR REPEAT-TRACK INSAR DATA PROCESSING Mingsheng LIAO, Li ZHANG, Zxn ZHANG, Jiangqing ZHANG Whan Technical University of Srveying and Mapping, Natinal Lab. for Information Eng. in

More information

A RECOGNITION METHOD FOR AIRPLANE TARGETS USING 3D POINT CLOUD DATA

A RECOGNITION METHOD FOR AIRPLANE TARGETS USING 3D POINT CLOUD DATA A RECOGNITION METHOD FOR AIRPLANE TARGETS USING 3D POINT CLOUD DATA Mei Zho*, Ling-li Tang, Chan-rong Li, Zhi Peng, Jing-mei Li Academy of Opto-Electronics, Chinese Academy of Sciences, No.9, Dengzhang

More information

The extra single-cycle adders

The extra single-cycle adders lticycle Datapath As an added bons, we can eliminate some of the etra hardware from the single-cycle path. We will restrict orselves to sing each fnctional nit once per cycle, jst like before. Bt since

More information

Multi-lingual Multi-media Information Retrieval System

Multi-lingual Multi-media Information Retrieval System Mlti-lingal Mlti-media Information Retrieval System Shoji Mizobchi, Sankon Lee, Fmihiko Kawano, Tsyoshi Kobayashi, Takahiro Komats Gradate School of Engineering, University of Tokshima 2-1 Minamijosanjima,

More information

Maximal Cliques in Unit Disk Graphs: Polynomial Approximation

Maximal Cliques in Unit Disk Graphs: Polynomial Approximation Maximal Cliqes in Unit Disk Graphs: Polynomial Approximation Rajarshi Gpta, Jean Walrand, Oliier Goldschmidt 2 Department of Electrical Engineering and Compter Science Uniersity of California, Berkeley,

More information

The single-cycle design from last time

The single-cycle design from last time lticycle path Last time we saw a single-cycle path and control nit for or simple IPS-based instrction set. A mlticycle processor fies some shortcomings in the single-cycle CPU. Faster instrctions are not

More information

What s New in AppSense Management Suite Version 7.0?

What s New in AppSense Management Suite Version 7.0? What s New in AMS V7.0 What s New in AppSense Management Site Version 7.0? AppSense Management Site Version 7.0 is the latest version of the AppSense prodct range and comprises three prodct components,

More information

Seismic trace interpolation with approximate message passing Navid Ghadermarzy and Felix Herrmann and Özgür Yılmaz, University of British Columbia

Seismic trace interpolation with approximate message passing Navid Ghadermarzy and Felix Herrmann and Özgür Yılmaz, University of British Columbia Seismic trace interpolation with approximate message passing Navid Ghadermarzy and Felix Herrmann and Özgür Yılmaz, University of British Colmbia SUMMARY Approximate message passing (AMP) is a comptationally

More information

Uncertainty Determination for Dimensional Measurements with Computed Tomography

Uncertainty Determination for Dimensional Measurements with Computed Tomography Uncertainty Determination for Dimensional Measrements with Compted Tomography Kim Kiekens 1,, Tan Ye 1,, Frank Welkenhyzen, Jean-Pierre Krth, Wim Dewlf 1, 1 Grop T even University College, KU even Association

More information

Designing and Optimization of Heterogeneous OTN/DWDM Networks with Intermediate Grooming

Designing and Optimization of Heterogeneous OTN/DWDM Networks with Intermediate Grooming Designing and Optimization of Heterogeneos OTN/DWDM Networks with Intermediate Grooming Afonso Oliveira IST (METI afonso.oliveira@ist.tl.pt ABSTRACT The coexistence of annels with different bitrates in

More information

CS 251, Winter 2018, Assignment % of course mark

CS 251, Winter 2018, Assignment % of course mark CS 25, Winter 28, Assignment 4.. 3% of corse mark De Wednesday, arch 7th, 4:3P Lates accepted ntil Thrsday arch 8th, am with a 5% penalty. (6 points) In the diagram below, the mlticycle compter from the

More information

EUCLIDEAN SKELETONS USING CLOSEST POINTS. Songting Luo. Leonidas J. Guibas. Hong-Kai Zhao. (Communicated by the associate editor name)

EUCLIDEAN SKELETONS USING CLOSEST POINTS. Songting Luo. Leonidas J. Guibas. Hong-Kai Zhao. (Communicated by the associate editor name) Volme X, No. 0X, 200X, X XX Web site: http://www.aimsciences.org EUCLIDEAN SKELETONS USING CLOSEST POINTS Songting Lo Department of Mathematics, University of California, Irvine Irvine, CA 92697-3875,

More information