Heuristic Algorithm for the Parallel Machine Total Weighted Tardiness Scheduling Problem


 Tyrone Adams
 1 years ago
 Views:
Transcription
1 Heuristic Algorithm for the Parallel Machine Total Weighted Tardiness Scheduling Problem Rosiane Rodrigues COPPE  Engenharia de Sistemas e Computação Universidade Federal do Rio de Janeiro Departamento de Ciência da Computação Universidade Federal do Amazonas Artur Pessoa, Eduardo Uchoa {artur, Departamento de Engenharia de Produção Universidade Federal Fluminense Marcus Poggi de Aragão Departamento de Informática Pontifícia Universidade Católica do Rio de Janeiro Abstract This paper presents a heuristic algorithm for the parallel machine weighted tardiness scheduling problem (P w j T j ). The main innovative feature of the algorithm is its representation of a multimachine schedule by a single sequence, greatly simplifying the treatment of that problem. The single sequence is optimized using an iterated local search over generalized pairwise interchange moves, improved with a suitable tie breaking criterion. Extensive tests on instances, with 2 and 4 machines, and with up to 50 jobs, obtained very good results, finding optimal solutions in almost all cases.
2 Relatórios de Pesquisa em Engenharia de Produção V. 8 n Introduction Let J = {1,..., n} be a set of jobs to be processed in a set of parallel identical machines M = {1,..., m} without preemption. Each machine can process at most one job at a time and each job must be processed by a single machine. Each job j has a positive processing time p j, a due date d j, and a positive weight w j. The tardiness of a job j with respect to its due date is defined as T j = max{0, C j d j }, where C j is the job completion time. The scheduling problem considered in this paper consists in sequencing the jobs in the machines in order to minimize n j=1 w jt j. This problem, referred as P w j T j in the 3field notation [10], is strongly NPhard, since even the singlemachine case (referred as 1 w j T j ) has that complexity [12]. On the other hand, the singlemachine problem without weights (1 T j ) is NPhard in the ordinary sense [7], but can solved in pseudopolynomial time [12]. All those problems may play an important role in real applications, such as in the manufacturing industry. We are not aware of other heuristic algorithms specially devised for the P w j T j. Recent works on related problems include Anghinolfi and Paolucci (2007) [2], that proposes a hybrid metaheuristic mixing Tabu Search, Simulated Annealing and Variable Neigborhood Search for the P T j ; and Bilge et al. (2004) [4], that proposes a Tabu Search for the P s ij T j (considering sequence dependent setup times). On the other hand, there is a rich literature on heuristics for the 1 w j T j. The most competitive heuristics for that problem are based on local searches that use both insertion moves (remove a job from the sequence and reinsert it in another position) and swap moves (swap the positions of a pair of jobs in the sequence). Those combined moves are also known as generalized pairwise interchange (GPI) moves [6]. Some 1 w j T j heuristics use the GPI moves in a traditional way, making one move at a time [20]. However, better results are obtained using dynamic programming to explore a large neighborhood that consists in certain sequences of single moves, the socalled dynasearch technique. While the size of the neighborhood is potentially exponential, the dynamic programming may determine such an optimal sequence of moves in polynomial time. This technique was introduced by Potts and Val de Verde [17] in the Traveling Salesman Problem (see the survey by Ahuja et al. [1]). Congram et al. [5] first applied dynasearch to the 1 w j T j, but only considering swaps as single moves. Grosso et al. [11] adapted the dynasearch technique in order to consider all GPI moves. Both papers present dynamic programming procedures with O(n 3 ) time complexity per search. Recently, Ergun and Orlin [9] improved this time complexity to O(n 2 ), by giving a faster dynasearch that combines GPI moves and also twist moves. Our heuristic algorithm for the P w j T j features the representation of a multimachine schedule by a single sequence, that can be improved by a local search using the wellknown GPI moves. Unfortunately, the single sequence representation does not appear to be suitable to dynasearch. This happens because this indirect representation makes the evaluation of moves to be much more costly, even a single insertion move may change the completion times of most jobs in a complex way, so the tardiness costs may have to be recomputed from scratch in O(n log m) time. Evaluating the tardiness costs of many sequences of moves is prohibitively expensive. In order to partially remedy the lack of a dynasearch, we introduce a tie breaking criterion in the GPI search. The idea is that even when the current solution has no GPI neighbor with better tardiness cost, the search is not stopped if there are neighbor solutions improving this criterion. The criterion was devised with the hope of guiding the search into solutions that more likely to be further improved. Several exact methods were proposed for the 1 w j T j, including [8, 19, 22, 21, 3, 14, 15]. The last three works describe techniques that can solve instances with up to 100 jobs consistently. However, the best heuristics typically find the optimal solutions for those instances in much less
3 Relatórios de Pesquisa em Engenharia de Produção V. 8 n time. By the way, those heuristics are often embedded as part of those branchandbound based exact algorithms. A good heuristic solution can reduce the search tree, leading to significant speedups. The only exact algorithm for the P w j T j that we know is proposed in [15]. It can solve most instances with up to 50 jobs. The heuristic presented in this paper was created in order to embedded in that exact algorithm. 2 The Algorithm A scheduling for the P w j T j is defined as a sequence of jobs on each machine, where each job appears in exactly one sequence. It is assumed that a machine is only idle after all jobs in its sequence are processed. Moreover, we say that a scheduling is minimal if no machine is idle before all jobs have started their processing. Of course, there is at least one optimal scheduling that is minimal. The main idea of our heuristic algorithm is to represent a minimal P w j T j scheduling as a single sequence of jobs. Let π = (π 1,..., π n ) be a permutation of the job indices 1,..., n. A minimal scheduling can be obtained from π as follows. Start with an empty schedule S, where all machines are idle at time 0. Then, for i = 1,..., n, schedule the job J πi in the first machine that becomes idle according to S. When ties occur, choose the machine with the smallest index. Denote the obtained schedule and its cost by S π and w(π), respectively. Note that every minimal schedule S can be generated in this way. To see this, construct π so that the sequence of jobs (J π1,..., J πn ) is in a nondecreasing order their starting times according to S. When ties occur, put the jobs scheduled in machines with smaller indices first. It is easy to see that S π = S. Given an initial permutation π of jobs, our algorithm performs a local search over a neighborhood defined by GPI moves. They are defined as either an exchange of two (not necessarily adjacent) jobs in π or the removal of a job from π followed by its insertion in another position. Each local search is performed until no GPI move improves the current solution. During the local search, a move is accepted only when the new solution improves upon the previous one. However, locally optimal solutions often have many neighbor solutions with the same cost. Hence, a premature stop in such local optima may be avoided using a suitable tie breaking criterion. Our criterion is based on the values of the due dates and the reverse position of the job in the sequence. In a formal way, given a sequence π = (π 1,..., π n ), we define b(π) = n d πj (n j + 1). Thus, given two neighbor sequences of jobs π and ρ with the same j=1 cost, if b(ρ) < b(π), then ρ is considered to improve over π. Note that our criterion induces the jobs having earlier due dates to be scheduled before the ones with later due dates. This criterion is motivated by the wellknown Earliest Due Date first (EDD) rule [13, 16], which generates an optimal solution for the 1 w j T j when the optimum value is zero (i.e., when there is a solution without tardy jobs). Algorithm 1 below presents the general steps of our heuristic, where N, r and k are parameters.
4 Relatórios de Pesquisa em Engenharia de Produção V. 8 n Algorithm 1 Single sequence based heuristic for the P w j T j i 1; π a permutation following the EDD rule. While i N If i is a multiple of r, then π a random permutation. Apply GPI moves in π, until no improvement is possible. If w(π) < w(π ), then π π. Apply k randomly chosen 2change moves in π. i i + 1. First, it generates a feasible solution following the EDD rule and store in π. In the first iteration, it generates a random permutation π, where the probabilities are identically distributed among all permutations. Then, GPI moves are applied to π. The search on the GPI neighborhood of the current π is stopped when the first improvement move is found, in this case π is updated and a new search starts. This is done until a search is completed without improvement (considering the tie breaking criterion). If the obtained solution improves upon the best solution found so far, then it is kept as π. Finally, k randomly chosen GPI moves are applied (regardless of whether they generate improvements or not), in an attempt to escape from bad local optimal regions. On every r iterations, a completely random permutation replaces the solution generated in the previous iteration. A complete search on the GPI neighborhood tests O(n 2 ) moves. The evaluation of each move requires the construction of the corresponding scheduling, this takes O(n log m) time. 3 Computational Experiments In all our experiments, we set N = 30mn, r = 5, and k = 3. obtained with a Intel Xeon 2.33 GHz processor. The reported times were 3.1 SingleMachine Even though our algorithm was devised for the P w j T j, we also tested it on 1 w j T j instances in order to benchmark it against other methods from the literature. The experiments were performed on the set of 375 instances of the problem available at the ORLibrary. This set was generated by Potts and Wassenhove [18] and contains 125 instances for each n {40, 50, 100}. In fact, for each such n, they created 5 similar instances (changing the seed) for each of 25 parameter configurations of the random instance generator. Therefore, for each value of n there are 25 groups composed by 5 similar instances. Those parameter configurations have influence on the distribution of the due dates. Processing times and weights are always picked from the discrete uniform distribution on [1,100] and [1,10], respectively. The best known results for those instances were obtained by Grosso et al. [11] using the GPIbased Dynasearch algorithm (GPIDS). In that paper, a comparative analysis was done between this method and an older method called AntColony Optimization (ACO), proposed by Stützle et al. [20], that uses the GPI moves in an ACO framework. Table 1 contains a comparison among our algorithm and the two methods above mentioned. The comparison is done on the time (average and maximum) needed to find the optimal value. The CPU times of
5 Relatórios de Pesquisa em Engenharia de Produção V. 8 n GPIDS were obtained in a HP Kayak 800 MHz workstation, while ACO was run on a Pentium III 450 MHz. Table 1: CPU time to find an optimal solution, GPIDS [11], ACO [20], and our algorithm, on OR Library instances with 40, 50 and 100 jobs (125 instances of each size). Topt,avg Topt,max GPIDS ACO Our method GPIDS ACO Our method GPIDS ACO Our method Although our heuristic (as GPIDS and ACO) eventually obtains the optimal solutions for all ORLibrary instances, our CPU times are significantly higher than those by GPIDS. However, even taking taking the difference of the processors into account, the performance of our heuristic is similar to ACO. This is quite remarkable, because when m = 1 our method reduces to a naive implementation of an iterated GPI local search. The factor that may explain why such a simple method matches the performance of a sophisticated metaheuristic using the same neighborhood is the improvement obtained by the tiebreaking criterion. In fact, without that criterion our method would fail to find the optimal solutions for 2 instances with 100 jobs, even after 3000 iterations and 160 seconds of CPU time. On the instances with 100 jobs, the average quality of the solutions found after each GPI local search iteration is 0.67% above the optimal, without the tiebreaking criteria this number would be 1.26% above the optimal. 3.2 MultiMachines In order to perform experiments on the P w j T j problem, we derived 100 new instances from the previously mentioned 1 w j T j ORLibrary instances. For m {2, 4}, n {40, 50}, we pick the first instance in each group (those with numbers ending with the digit 1 or 6) and divided each due date d j by m (and rounded down the result), processing times p j and weights w j are kept unchanged. For example, from instance wt401, we produced instances wt402m1 and wt404m1 by dividing due dates by 2 and 4, respectively. The exact algorithm presented in [15] solved to optimality 98 out of those 100 instances, so we have a good basis to assess the performance of our heuristic. Tables 2 to 5 present detailed results of our heuristic algorithm. The columns have the following meaning: (1) the instance number; (2) the optimal solution value; (3) the value of the best solution found by the heuristic; (4) the difference between the previous values; (5) the iteration in which the best value was first found; (6) the elapsed CPU time (in seconds) when the best value was first found; (7) the number of iterations where the best value was found; (8) the average value of the solutions obtained in all iterations; (9) the total number of iterations performed (given by the algorithm parameter N); and, (10) the total CPU time (in seconds). Table 2 and Table 3 present detailed results considering 2 machines, for 40 and 50 jobs, respectively. Only for a single instance (wt402m116), the optimal solution (or the best known solution, in case of instance wt502m31) was not found. Table 4 and Table 5 present detailed
6 Relatórios de Pesquisa em Engenharia de Produção V. 8 n results considering 4 machines, for 40 and 50 jobs, respectively. Optimal solutions (or the best known solution, in case of instance wt504m56) were found for all tested instances. We also run all instances with a simpler variant of our method, without the tiebreaking criterion. Table 6 compares statistics on those runs with statistics on the runs using the complete method, with the tiebreaking criterion. For each value of m and n, it shows: (1) the number of instances where the optimal or best known solution value was not found; (2) the average quality of the solutions found after each GPI local search iteration; and (3) the total time to complete all the N iterations. It can be seen that the tiebreaking criterion makes the algorithm significantly more robust, with a modest increase on the time spent per iteration (28% in average).
7 Relatórios de Pesquisa em Engenharia de Produção V. 8 n Table 2: Results on instances with 40 jobs and 2 machines. Inst Opt BestV IterBest TBest(s) #Best AvgV TotIter TotT(s) Avg :
8 Relatórios de Pesquisa em Engenharia de Produção V. 8 n Table 3: Results on instances with 50 jobs and 2 machines. Inst Opt BestV IterBest TBest(s) #Best AvgV TotIter TotT(s) Avg :
9 Relatórios de Pesquisa em Engenharia de Produção V. 8 n Table 4: Results on instances with 40 jobs and 4 machines. Inst Opt BestV IterBest TBest(s) #Best AvgV TotIter TotT(s) Avg :
10 Relatórios de Pesquisa em Engenharia de Produção V. 8 n Table 5: Results on instances with 50 jobs and 4 machines. Inst Opt BestV IterBest TBest(s) #Best AvgV TotIter TotT(s) Avg :
11 Relatórios de Pesquisa em Engenharia de Produção V. 8 n Table 6: Effect of the tiebreaking criterion on the proposed algorithm Without tiebreaking With tiebreaking m n #N onopt AvgQ(%) AvgTotT(s) #N onopt AvgQ(%) AvgTotT(s) References [1] R. Ahuja, O. Ergun, J. Orlin, and A. Punnen. A survey of verylarge scale neighborhood search techniques. Discrete Applied Mathematics, 123:75 103, [2] D. Anghinolfi and M. Paolucci. Parallel machine total tardiness scheduling with a new hybrid metaheuristic approach. Computers and Operations Research, 34: , [3] L. Bigras, M. Gamache, and G. Savard. Timeindexed formulations and the total weighted tardiness problem. INFORMS Journal on Computing, 1: , [4] U. Bilge, F. Kyraç, F. Kurtulan, and M. Pekgun. A tabu search algorithm for parallel machine total tardiness problem. Computers and Operations Research, 31: , [5] R. Congram, C. Potts, and S. van de Velde. An iterated dynasearch algorithm for the single machine total weighted tardiness scheduling problem. INFORMS Journal of Computing, 14:52 67, [6] F. Della Croce. Generalized pairwise interchanges and machine scheduling. European Journal of Operational Research, 83: , [7] J. Du and J. Leung. Minimizing total tardiness on one processor is NPHard. Mathematics of Operations Research, 15(3): , [8] M. Dyer and L. Wolsey. Formulating the single machine sequencing problem with release dates as a mixed integer program. Discrete Applied Mathematics, 26: , [9] O. Ergun and J. Orlin. Fast neighborhood search for the single machine total weighted tardiness problem. Operations Research Letters, 34:41 45, [10] R. Graham, E. Lawler, J. Lenstra, and A. RinnooyKan. Optimization and approximation in deterministic sequencing and scheduling: a survey. Annals of Discrete Mathematics, 5: , [11] A. Grosso, F. Della Croce, and R. Tadei. An enhanced dynasearch neighborhood for the singlemachine total weighted tardiness scheduling problem. Operations Research Letters, 32:68 72, [12] E. Lawler. A pseudopolynomial algorithm for sequencing jobs to minimize total tardiness. Annals of Research Letters, 1: , [13] J. Leung. Handbook of Scheduling: Algorithms, Models, and Performance Analysis. Chapman&Hall and CRS Press, 2004.
12 Relatórios de Pesquisa em Engenharia de Produção V. 8 n [14] Y. Pan and L. Shi. On the equivalence of the maxmin transportation lower bound and the timeindexed lower bound for single machine scheduling problems. Mathematical Programming, 110: , [15] A. Pessoa, E. Uchoa, M. Poggi de Aragão, and R. Rodrigues. Algorithms over arctime indexed formulations for single and parallel machine scheduling problems. Technical Report RPEP Vol.8 no.8, Universidade Federal Fluminense, Engenharia de Produção, Niterói, Brazil, [16] M. Pinedo. Scheduling: theory, algorithms, and systems. PrenticeHall, [17] C. Potts and S. van de Velde. Dynasearch  iterative local improvement by dynamic programming. Part I: the traveling salesman problem. Technical report, University of Twente, [18] C. Potts and L. Wassenhove. A branchandbound algorithm for the total weighted tardiness problem. Operations Research, 32: , [19] J. Sousa and L. Wolsey. A time indexed formulation of nonpreemptive single machine scheduling problems. Mathematical Programming, 54: , [20] T. Stützle, M. Den Besten, and M. Dorigo. Ant colony optimization for the total weighted tardiness problem. Technical report, Technical Report IRIDIA/9916, [21] J. Van der Akker, C. Hurkens, and M. Savelsbergh. Timeindexed formulations for machine scheduling problems:column generation. INFORMS Journal on Computing, 12(2): , [22] J. Van der Akker, C. Van Hoesel, and M. Savelsbergh. A polyhedral approach to singlemachine scheduling problems. Mathematical Programming, 85: , 1999.
RELAXATION HEURISTICS FOR THE SET COVERING PROBLEM
Journal of the Operations Research Society of Japan 2007, Vol. 50, No. 4, 350375 RELAXATION HEURISTICS FOR THE SET COVERING PROBLEM Shunji Umetani University of ElectroCommunications Mutsunori Yagiura
More informationSizeConstrained Weighted Set Cover
SizeConstrained Weighted Set Cover Lukasz Golab 1, Flip Korn 2, Feng Li 3, arna Saha 4 and Divesh Srivastava 5 1 University of Waterloo, Canada, lgolab@uwaterloo.ca 2 Google Research, flip@google.com
More informationAnt Colony System: A Cooperative Learning Approach to the Traveling Salesman Problem
Ant Colony System: A Cooperative Learning Approach to the Traveling Salesman Problem TR/IRIDIA/19965 Université Libre de Bruxelles Belgium Marco Dorigo IRIDIA, Université Libre de Bruxelles, CP 194/6,
More informationDealing with Uncertainty in Operational Transport Planning
Dealing with Uncertainty in Operational Transport Planning Jonne Zutt, Arjan van Gemund, Mathijs de Weerdt, and Cees Witteveen Abstract An important problem in transportation is how to ensure efficient
More informationMaximizing the Spread of Influence through a Social Network
Maximizing the Spread of Influence through a Social Network David Kempe Dept. of Computer Science Cornell University, Ithaca NY kempe@cs.cornell.edu Jon Kleinberg Dept. of Computer Science Cornell University,
More informationSubspace Pursuit for Compressive Sensing: Closing the Gap Between Performance and Complexity
Subspace Pursuit for Compressive Sensing: Closing the Gap Between Performance and Complexity Wei Dai and Olgica Milenkovic Department of Electrical and Computer Engineering University of Illinois at UrbanaChampaign
More informationOPRE 6201 : 2. Simplex Method
OPRE 6201 : 2. Simplex Method 1 The Graphical Method: An Example Consider the following linear program: Max 4x 1 +3x 2 Subject to: 2x 1 +3x 2 6 (1) 3x 1 +2x 2 3 (2) 2x 2 5 (3) 2x 1 +x 2 4 (4) x 1, x 2
More informationDistributed Optimization by Ant Colonies
APPEARED IN PROCEEDINGS OF ECAL91  EUROPEAN CONFERENCE ON ARTIFICIAL LIFE, PARIS, FRANCE, ELSEVIER PUBLISHING, 134 142. Distributed Optimization by Ant Colonies Alberto Colorni, Marco Dorigo, Vittorio
More informationScalable Collaborative Filtering with Jointly Derived Neighborhood Interpolation Weights
Seventh IEEE International Conference on Data Mining Scalable Collaborative Filtering with Jointly Derived Neighborhood Interpolation Weights Robert M. Bell and Yehuda Koren AT&T Labs Research 180 Park
More informationRealTime Dynamic Voltage Scaling for LowPower Embedded Operating Systems
RealTime Dynamic Voltage Scaling for LowPower Embedded Operating Syste Padmanabhan Pillai and Kang G. Shin RealTime Computing Laboratory Department of Electrical Engineering and Computer Science The
More informationCLoud Computing is the long dreamed vision of
1 Enabling Secure and Efficient Ranked Keyword Search over Outsourced Cloud Data Cong Wang, Student Member, IEEE, Ning Cao, Student Member, IEEE, Kui Ren, Senior Member, IEEE, Wenjing Lou, Senior Member,
More informationLearning to Select Features using their Properties
Journal of Machine Learning Research 9 (2008) 23492376 Submitted 8/06; Revised 1/08; Published 10/08 Learning to Select Features using their Properties Eyal Krupka Amir Navot Naftali Tishby School of
More informationRobust Set Reconciliation
Robust Set Reconciliation Di Chen 1 Christian Konrad 2 Ke Yi 1 Wei Yu 3 Qin Zhang 4 1 Hong Kong University of Science and Technology, Hong Kong, China 2 Reykjavik University, Reykjavik, Iceland 3 Aarhus
More informationThe Set Covering Machine
Journal of Machine Learning Research 3 (2002) 723746 Submitted 12/01; Published 12/02 The Set Covering Machine Mario Marchand School of Information Technology and Engineering University of Ottawa Ottawa,
More informationOn SetBased Multiobjective Optimization
1 On SetBased Multiobjective Optimization Eckart Zitzler, Lothar Thiele, and Johannes Bader Abstract Assuming that evolutionary multiobjective optimization (EMO) mainly deals with set problems, one can
More informationIEEE TRANSACTIONS ON AUDIO, SPEECH, AND LANGUAGE PROCESSING, 2013. ACCEPTED FOR PUBLICATION 1
IEEE TRANSACTIONS ON AUDIO, SPEECH, AND LANGUAGE PROCESSING, 2013. ACCEPTED FOR PUBLICATION 1 ActiveSet Newton Algorithm for Overcomplete NonNegative Representations of Audio Tuomas Virtanen, Member,
More informationAn Experimental Comparison of MinCut/MaxFlow Algorithms for Energy Minimization in Vision
In IEEE Transactions on PAMI, Vol. 26, No. 9, pp. 11241137, Sept. 2004 p.1 An Experimental Comparison of MinCut/MaxFlow Algorithms for Energy Minimization in Vision Yuri Boykov and Vladimir Kolmogorov
More informationBetween Restarts and Backjumps
Between Restarts and Backjumps Antonio Ramos, Peter van der Tak, and Marijn Heule Department of Software Technology, Delft University of Technology, The Netherlands Abstract. This paper introduces a novel
More informationThe Set Data Model CHAPTER 7. 7.1 What This Chapter Is About
CHAPTER 7 The Set Data Model The set is the most fundamental data model of mathematics. Every concept in mathematics, from trees to real numbers, is expressible as a special kind of set. In this book,
More informationSo Who Won? Dynamic Max Discovery with the Crowd
So Who Won? Dynamic Max Discovery with the Crowd Stephen Guo Stanford University Stanford, CA, USA sdguo@cs.stanford.edu Aditya Parameswaran Stanford University Stanford, CA, USA adityagp@cs.stanford.edu
More informationNearOptimal Sensor Placements in Gaussian Processes
Carlos Guestrin Andreas Krause Ajit Paul Singh School of Computer Science, Carnegie Mellon University GUESTRIN@CS.CMU.EDU KRAUSEA@CS.CMU.EDU AJIT@CS.CMU.EDU Abstract When monitoring spatial phenomena,
More informationCommunication Theory of Secrecy Systems
Communication Theory of Secrecy Systems By C. E. SHANNON 1 INTRODUCTION AND SUMMARY The problems of cryptography and secrecy systems furnish an interesting application of communication theory 1. In this
More informationGiotto: A TimeTriggered Language for Embedded Programming
Giotto: A TimeTriggered Language for Embedded Programming THOMAS A HENZINGER, MEMBER, IEEE, BENJAMIN HOROWITZ, MEMBER, IEEE, AND CHRISTOPH M KIRSCH Invited Paper Giotto provides an abstract programmer
More informationApproximately Detecting Duplicates for Streaming Data using Stable Bloom Filters
Approximately Detecting Duplicates for Streaming Data using Stable Bloom Filters Fan Deng University of Alberta fandeng@cs.ualberta.ca Davood Rafiei University of Alberta drafiei@cs.ualberta.ca ABSTRACT
More informationPreferencebased Search using ExampleCritiquing with Suggestions
Journal of Artificial Intelligence Research 27 (2006) 465503 Submitted 04/06; published 12/06 Preferencebased Search using ExampleCritiquing with Suggestions Paolo Viappiani Boi Faltings Artificial
More informationEfficient Processing of Joins on Setvalued Attributes
Efficient Processing of Joins on Setvalued Attributes Nikos Mamoulis Department of Computer Science and Information Systems University of Hong Kong Pokfulam Road Hong Kong nikos@csis.hku.hk Abstract Objectoriented
More informationIntroduction to partitioningbased clustering methods with a robust example
Reports of the Department of Mathematical Information Technology Series C. Software and Computational Engineering No. C. 1/2006 Introduction to partitioningbased clustering methods with a robust example
More informationFast Greeks by algorithmic differentiation
The Journal of Computational Finance (3 35) Volume 14/Number 3, Spring 2011 Fast Greeks by algorithmic differentiation Luca Capriotti Quantitative Strategies, Investment Banking Division, Credit Suisse
More informationEfficient Algorithms for Sorting and Synchronization Andrew Tridgell
Efficient Algorithms for Sorting and Synchronization Andrew Tridgell A thesis submitted for the degree of Doctor of Philosophy at The Australian National University February 1999 Except where otherwise
More informationHow to Use Expert Advice
NICOLÒ CESABIANCHI Università di Milano, Milan, Italy YOAV FREUND AT&T Labs, Florham Park, New Jersey DAVID HAUSSLER AND DAVID P. HELMBOLD University of California, Santa Cruz, Santa Cruz, California
More information