Traineeship proposal ANALYSIS AND COMPARISON OF PARAMETER TUNING FOR LOCAL SEARCH ALGORITHMSLocation either University of Nantes LINA or University of Angers LERIA FranceSalary up to for the whole periodStarting date May Duration at least monthsContact Charlotte Truchet charlotte nantes fr and Frédéric Saubion Frederic angers frContextDuring the last decades impressive improvements have been achieved to solve complexoptimization problems issued from real world applications which involve more and more dataand constraints In order to tackle large scale instances and intricate problem structures sophisticated solving techniques have been developed and combined to provide efficientsolvers Among the different solving paradigms local search has been widely used as an incompleteoptimization technique for solving such problems It is now integrated in solvers and combinedwith other techniques Local search mainly relies on the basic concept of neighbourhood Starting from an initialconfiguration a local search algorithm tries to reach the optimum by moving locally from aconfiguration to one of its neighbours according to its evaluation The performance of such analgorithm is strongly related to its ability to explore and exploit the search landscape Forinstance when faced to a very rugged landscape one should be able to escape from manylocal optima while in presence of large plateaus one should be able to widely explore thespace In order to manage the balance between exploitation and exploration various efficientheuristics have been proposed usually relying on stochastic perturbations and restarts Unfortunately these heuristics are most of the time controlled by parameters whose settinghas a great impact on the efficiency of the algorithm Well known parameters are for instancethe temperature cooling schedule in simulated annealing or the amount of random walk Parameter tuning is nowadays a crucial issue and various tuning methods have beendeveloped including a dynamic management of parameters during the solving process ...
3
pages
English
Documents
Obtenez un accès à la bibliothèque pour le consulter en ligne En savoir plus
Découvre YouScribe et accède à tout notre catalogue !
Découvre YouScribe et accède à tout notre catalogue !
3
pages
English
Documents
Obtenez un accès à la bibliothèque pour le consulter en ligne En savoir plus
Niveau: Supérieur, Doctorat, Bac+8
Traineeship proposal ANALYSIS AND COMPARISON OF PARAMETER TUNING FOR LOCAL SEARCH ALGORITHMSLocation : either University of Nantes (LINA) or University of Angers (LERIA), FranceSalary : up to 3000 € for the whole periodStarting date : May 2010 Duration : at least 4 monthsContact : Charlotte Truchet, , and Frédéric Saubion,ntextDuring the last decades, impressive improvements have been achieved to solve complexoptimization problems, issued from real world applications, which involve more and more dataand constraints. In order to tackle large scale instances and intricate problem structures,sophisticated solving techniques have been developed and combined to provide efficientsolvers. Among the different solving paradigms, local search has been widely used as an incompleteoptimization technique for solving such problems. It is now integrated in solvers and combinedwith other techniques. Local search mainly relies on the basic concept of neighbourhood. Starting from an initialconfiguration, a local search algorithm tries to reach the optimum by moving locally from aconfiguration to one of its neighbours, according to its evaluation. The performance of such analgorithm is strongly related to its ability to explore and exploit the search landscape. Forinstance, when faced to a very rugged landscape, one should be able to escape from manylocal optima while in presence of large plateaus, one should be able to widely explore thespace.In order to manage the balance between exploitation and exploration, various efficientheuristics have been proposed, usually relying on stochastic perturbations and restarts.
Voir
Traineeship proposal ANALYSIS AND COMPARISON OF PARAMETER TUNING FOR LOCAL SEARCH ALGORITHMSLocation : either University of Nantes (LINA) or University of Angers (LERIA), FranceSalary : up to 3000 € for the whole periodStarting date : May 2010 Duration : at least 4 monthsContact : Charlotte Truchet, , and Frédéric Saubion,ntextDuring the last decades, impressive improvements have been achieved to solve complexoptimization problems, issued from real world applications, which involve more and more dataand constraints. In order to tackle large scale instances and intricate problem structures,sophisticated solving techniques have been developed and combined to provide efficientsolvers. Among the different solving paradigms, local search has been widely used as an incompleteoptimization technique for solving such problems. It is now integrated in solvers and combinedwith other techniques. Local search mainly relies on the basic concept of neighbourhood. Starting from an initialconfiguration, a local search algorithm tries to reach the optimum by moving locally from aconfiguration to one of its neighbours, according to its evaluation. The performance of such analgorithm is strongly related to its ability to explore and exploit the search landscape. Forinstance, when faced to a very rugged landscape, one should be able to escape from manylocal optima while in presence of large plateaus, one should be able to widely explore thespace.In order to manage the balance between exploitation and exploration, various efficientheuristics have been proposed, usually relying on stochastic perturbations and restarts.
- evolutionary computation
- experimental dataconsidering random
- well known parameters
- proceedings des journées francophones de programmation par contraintes
- extensive experimental studies
Traineeship proposal
ANALYSISANDCOMPARISONOFPARAMETERTUNINGFORLOCALSEARCHALGORITHMS
Location : either University of Nantes (LINA) or University of Angers (LERIA), France
Salary : up to 3000€for the whole period
Starting date : May 2010
Duration : at least 4 months
Contact : Charlotte Truchet,charlotte.truchet univ-nantes.fr, and Frédéric Saubion, Frederic.Saubion@univ-angers.fr Context During the last decades, impressive improvements have been achieved to solve complex optimization problems, issued from real world applications, which involve more and more data and constraints. In order to tackle large scale instances and intricate problem structures, sophisticated solving techniques have been developed and combined to provide efficient solvers.
Among the different solving paradigms, local search has been widely used as an incomplete optimization technique for solving such problems. It is now integrated in solvers and combined with other techniques.
Local search mainly relies on the basic concept of neighbourhood. Starting from an initial configuration, a local search algorithm tries to reach the optimum by moving locally from a configuration to one of its neighbours, according to its evaluation. The performance of such an algorithm is strongly related to its ability to explore and exploit the search landscape. For instance, when faced to a very rugged landscape, one should be able to escape from many local optima while in presence of large plateaus, one should be able to widely explore the space. In order to manage the balance between exploitation and exploration, various efficient heuristics have been proposed, usually relying on stochastic perturbations and restarts. Unfortunately, these heuristics are most of the time controlled by parameters whose setting has a great impact on the efficiency of the algorithm. Well known parameters are for instance the temperature cooling schedule in simulated annealing or the amount of random walk. Parameter tuning is nowadays a crucial issue and various tuning methods have been developed, including a dynamic management of parameters during the solving process. Work plan
