Hart, W. E. (1994). Adaptive global optimization with local search. Doctoral dissertation, University of California, San Diego, San Diego, CA.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....algorithm when individuals were improved by 29local search with a probability 0.05 (i.e. when the local search probability LS p was specified as = LS p 0.05) Goldberg and Voessner [45] presented a theoretical framework for discussing the balance between genetic search and local search. Hart [47] investigated the following four questions for designing efficient memetic algorithms for continuous optimization: a) How often should local search be applied (b) On which solutions should local search be used (c) How long should local search be run (d) How efficient does local search need ....

....parameters is a future research topic. Tan et al. 49] proposed an idea of adjusting the number of solutions examined in local search in their multiobjective memetic algorithm. Many issues related to dynamic parameter control have already been studied for single objective memetic algorithms [34] [47], 48] 50] 51] Those studies can be extended to the case of multiobjective memetic algorithms where more emphasis should be placed on the diversity of solutions than the case of single objective optimization. The performance evaluation of our MOGLS in this paper is not complete. We compared ....

W. E. Hart, "Adaptive global optimization with local search," Ph. D. Thesis, University of -54California, San Diego, 1994.

....source code project. The les composing the Memetic Algorithm Framework can be seen in Fig. 1. 1 Introduction It is now well established that a combination of Genetic Algorithms with local search are amongst the most powerful metaheuristcs to search complex continuous or combinatorial spaces [5, 8]. GAs combined with local search (LS) were named Memetic Algorithms (MAs) in [10] In the literature, MAs have also been named Hybrid Genetic Algorithms, Genetic Local Searchers, Lamarckian Genetic Algorithms, Baldwinian Genetic Algorithms and even Parallel Genetic Algorithms. The goal of this ....

W. E. Hart. Adaptive global optimization with local search. Ph.D. Thesis, University of California, San Diego, 1994.

....evolution process. Although they have been demonstrated to be effective in many practical problems, their efficiencies are usually low and the algorithms often take a long time to converge. Many variants have been tried to combine them with some local search methods to improve their efficiencies. [9] No Free Lunch theorem[10] has demonstrated that no single algorithm can consistently perform better 2 in all classes of problems than the other algorithms. That is, for one class of problems where an algorithm can do better than other algorithms, there is always another class where it will do ....

W. E. Hart. Adaptive Global Optimization with Local Search. PhD thesis, University of California, San Diego, 1994.

....and natural evolution process. Although they have been successfully used in many practical problems, their efficiencies are usually low and the algorithms often take a long time to converge. Many variants have been attempted to combine them with some local search methods to improve efficiency. [7] The No Free Lunch theorem[8] has demonstrated that no single algorithm can consistently perform better in every class of problems than the other algorithms. That is, for one class of problems where an algorithm can do better than other algorithms, there is always another class where it will ....

W. E. Hart. Adaptive Global Optimization with Local Search. PhD thesis, University of California, San Diego, 1994. 18

....and situations, Memetic Algorithms are also known as Hybrid GAs, Genetic Local Searchers, Baldwinian GAs, Lamarkian GAs, etc. From an optimization point of view MAs have shown that they are orders of magnitude more accurate than traditional GAs for some problem domains. See for example reference [11] for a continuous domain research and [18] for combinatorial optimization studies. It is argued that the success of MAs is due to the tradeo between the exploration abilities of the underlying GA and the exploitation abilities of the local searchers used. The price to be paid is a greater number ....

W. E. Hart. Adaptive global optimization with local search. Ph.D. Thesis, University of California, San Diego, 1994.

.... University, Lawrence Livermore National Laboratories, Sandia National Laboratories, and the Cornell Theory Center for applications in gas dynamics [141] smoothed particle hydrodynamics, particle simulation studies [68] adaptive eigenvalue solvers in materials design [37, 94] genetics algorithms [81], adaptive multigrid methods in numerical relativity, and the dynamics of earthquake faults (see Section 6.3 for a complete list) Our parallel software infrastructure addresses two goals of software support for scientific applications [112] ffl it hides low level details of the hardware, and ....

....if the asynchronous message and the barrier message arrived out of order. The cost of this synchronization protocol is discussed in Section 3.3.3. An AMS Example Figure 3. 4 illustrates sample AMS C code based on a geographically structured genetics algorithm application developed using LPARX [81]. Recall from Chapter 2 that LPARX Grids may contain elements of any user defined type. In this particular application, Grid elements are of type GA Individual. For many user defined structures, such as those containing pointers or other user defined types, the LPARX run time system does not know ....

[Article contains additional citation context not shown here]

W. E. Hart, Adaptive Global Optimization with Local Search, PhD thesis, University of California at San Diego, 1994.

....and utilization in di erent problem domains. MAFRA is an open source code project. 1 Introduction It is now well established that a combination of Genetic Algorithms with local search are amongst the most powerful metaheuristcs to search complex continuous or combinatorial spaces (e.g. see [5, 8]) GAs combined with local search (LS) were named Memetic Algorithms (MAs) in [10] In the literature, MAs have also been named Hybrid Genetic Algorithms, Genetic Local Searchers, Lamarckian Genetic Algorithms, Baldwinian Genetic Algorithms and even Parallel Genetic Algorithms. We will use the ....

W. E. Hart. Adaptive global optimization with local search. Ph.D. Thesis, University of California, San Diego, 1994.

....algorithm grows exponentially with the number of nodes in the networks. Another global optimization algorithm [27] uses a user defined trace function to lead the search away from local optima, but it is doubtful that a suitable trace can easily be found. Gradient based algorithms with multistarts [12] can also be used, but the appropriate number of restarts is difficult to derive a priori and the computation time can be very long. It is widely believed that evolutionary search (see [3] 10] 32] and [37] for a review) and simulated annealing [7] can overcome the above difficulties, but these ....

....When the complexity is low, it is possible to apply local search at every generation, leading to the lifetime learning approach. Another factor that affects the frequency of applying local search is the evolutionary search s ability to eliminate the regions not containing the global optimum. Hart [12] found that the frequency should be reduced when the fitness distribution of the population reliably indicates the possible locations of the global optimum; otherwise, little benefit can be obtained from local search. Between the two local search methods that we have investigated, the delta rule ....

W. E. Hart. Adaptive Global Optimization with Local Search. PhD thesis, Department of Computer Science and Engineering, University of California, San Diego, 1994.

.... guarantee the uniformity of searching done by the GWW method [ L(ph) L g M g g f(ph) L L d(g) ph Ph G d 1 (g) Figure 2: Search in genotypic and phenotypic spaces ror critierion) prior to the assessment of their fitness, the result is a picture something like Figure 2 [Hart et al. 1994, Belew and Mitchell, 1996] This perspective generates a number of interesting issues, for example the conditions under which Lamarckian inheritance of acquired characters found by the local search do or do not speed the search by evolution. One immediate but significant computational ....

.... in the case of continuous problem spaces, where guarantees of local search methods within local neighborhoods are extended to seek solutions which are globally optimal across an entire domain, even when explicit gradient information is not available [Belew et al. 1991, Whitley and Hanson, 1989, Hart, 1994, Rosin et al. 1997, Land et al. 1997] In the case of discrete, combinatorial optimization definitions of local search neighborhoods must be in terms of operators which perturb one solution slightly to form alternatives. Classic examples include single bit changes performed by simulated ....

Hart, W. E. (1994). Adaptive global optimization with local search. PhD thesis, Computer Science & Engr. Dept. - Univ. Calif. San Diego.

....and Fortran provides highly optimized and vectorized numerical kernels. Furthermore, programmers may not be required to rewrite highly optimized Fortran kernels when parallelizing an application. Some current LPARX applications include a 2d geographically structured genetic algorithm application [20], a 3d smoothed particle hydrodynamics method [24] an adaptive eigenvalue solver for the first principles simulation of real materials [11, 24] and a dimension independent code for 2d, 3d, and 4d connected component labeling for spin models in statistical mechanics [18] In the following ....

W. E. Hart, Adaptive Global Optimization with Local Search, PhD thesis, University of California at San Diego, 1994.

....whole population. Therefore, we have adopted the approach similar to that of [9] 17] where learning was applied to fine tune every chromosome generated in each cycle of GAs. In our experiments, we have also investigated the effect of varying the learning frequency on the evolution process, as in [18]. Usually, learning methods depend very much on the chromosomal representation. For floating point representation, some researchers [23] 35] 41] used gradient descent algorithms such as backpropagation or its variants as the learning methods. In this case, the gradient in the fitness surface ....

W. E. Hart. Adaptive Global Optimization with Local Search. PhD thesis, Department of Computer Science and Engineering, University of California, San Diego, 1994.

....further research for this approach. In multi commodity fixed charge networks, for example, the subproblems would be single commodity subproblems corresponding to various approximations of the the fixedcharges. Recent results in telecommunications network design [Dav97] and pharmaceutical design [Har94] demonstrate that this paradigm of utilizing subproblem solutions as building blocks within the context of genetic algorithms is very effective in those problem domains as well. By focusing on the island GA model with appropriately designed high level problem representations to reduce ....

W. E. Hart. Adaptive global optimization with local search. PhD thesis, University of.California, San Diego, 1994.

....slow in fine tuning solutions. A local search method like back propagation (BP) which is used here is much faster at fine tuning [7] Hence, one wants to combine the advantages of both: genetic and local search. A good overview of past research on this combination can, for example, be found in [5]. The biological study of the interaction between life time learning and evolutionary learning has a long history. Molecular biology has firmly rejected the Lamarckian idea that individual adaptation (life time learning) can alter information in an individual s gametes. Or, in other words, the ....

....not only depends on the problem to be solved it also depends on the type of LTL which is used. Some interesting questions are: How much LTL should there be for a given problem What is the right size of (weight) changes resulting from LTL Should one use Lamarckian or Baldwinian learning Hart [5] has investigated some of these issues and their interaction for different types of single population GAs. His work clearly shows the complexity of the interactions between evolutionary learning and LTL. Hence, we suggest to let evolution itself solve these questions (as was done in nature) Our ....

Hart, W. E. H., (1994), Adaptive Global Optimization with Local Search, PhD Dissertation University of California, San Diego.

....Kan Timmer, 1984; Torn Zilinskas, 1989) use a Monte Carlo sample generation technique. Cluster analysis algorithms are used to identify local minima. This is followed by a local search for each local optimum. Clustering methods have been found useful for many global optimization problems (Hart, 1994; Torn Zilinskas, 1989) However they are likely to perform poorly when the objective function is multimodal and there are many local optima (Hart, 1994) Since the early 80s, the simulated annealing (SA) algorithms (Kirpatrick, Gelatt, Vecchi, 1983) and their variants have been used for ....

....This is followed by a local search for each local optimum. Clustering methods have been found useful for many global optimization problems (Hart, 1994; Torn Zilinskas, 1989) However they are likely to perform poorly when the objective function is multimodal and there are many local optima (Hart, 1994). Since the early 80s, the simulated annealing (SA) algorithms (Kirpatrick, Gelatt, Vecchi, 1983) and their variants have been used for solving blackbox problems. The natural motivation behind SA is the statistical behavior of molecules during the crystallization process in annealing. SA ....

Hart, W. E. (1994). Adaptive global optimization with local search. Doctoral dissertation, Department of Computer Science, University of California, San Diego.

.... perform a more sophisticated search across multiple local minima and are good at finding regions of the search space where nearoptimal solutions can be found; however, they are usually not as good at refining a solution (once it is close to a near optimal solution) as local optimization strategies (Hart, 1994). Recent research has shown that it is desirable to emply both a global and local search strategy (Hart, 1994) Hybrid GAs (such as Regent) combine local search with a more traditional GA. While we focus on hybrid GA algorithms in this section, this two tiered search strategy has been employed by ....

.... space where nearoptimal solutions can be found; however, they are usually not as good at refining a solution (once it is close to a near optimal solution) as local optimization strategies (Hart, 1994) Recent research has shown that it is desirable to emply both a global and local search strategy (Hart, 1994). Hybrid GAs (such as Regent) combine local search with a more traditional GA. While we focus on hybrid GA algorithms in this section, this two tiered search strategy has been employed by other researchers as well (Kohavi John, 1997; Provost Buchanan, 1995; Schaffer, 1993) GAs have been ....

[Article contains additional citation context not shown here]

Hart, W. (1994). Adaptive Global Optimization with Local Search. Ph.D. thesis, University of California, San Diego.

.... with adaptive sampling algorithms are which solutions (here ants representing the solutions) should be allowed to perform a local search (see [33] for a first discussion of this issue for the application of MAX MIN Ant System to the TSP) and how powerful the local search method should be [19, 37]. In this paper we present an extension of MAX MIN Ant System using a modified selection rule for the construction of solutions, the pseudo random proportional rule. This selection rule was first proposed for Ant Q [14] and is also used in Ant Colony System [15] the successor of Ant Q. The ....

W.E. Hart. Adaptive Global Optimization with Local Search. PhD thesis, University of California, San Diego, 1994.

....Kan Timmer, 1984; Torn Zilinskas, 1989) use a Monte Carlo sample generation technique. Cluster analysis algorithms are used to identify local minima. This is followed by a local search for each local optimum. Clustering methods have been found useful for many global optimization problems (Hart, 1994; Torn Zilinskas, 1989) However they are likely to perform poorly when the objective function is multimodal and there are many local optima (Hart, 1994) Since the early 80s, the simulated annealing (SA) algorithms (Kirpatrick, Gelatt, Vecchi, 1983) and their variants have been used for ....

....This is followed by a local search for each local optimum. Clustering methods have been found useful for many global optimization problems (Hart, 1994; Torn Zilinskas, 1989) However they are likely to perform poorly when the objective function is multimodal and there are many local optima (Hart, 1994). Since the early 80s, the simulated annealing (SA) algorithms (Kirpatrick, Gelatt, Vecchi, 1983) and their variants have been used for solving blackbox problems. The natural motivation behind SA is the statistical behavior of molecules during the crystallization process in annealing. SA ....

Hart, W. E. (1994). Adaptive global optimization with local search. Doctoral dissertation, Department of Computer Science, University of California, San Diego.

....of an individual probabilistically. This can be modeled by either (a) adding the value of a random variable to a dimension of an individual, called additive mutation, or (b) replacing a dimension of an individual with the value of a random variable, called replacement mutation. For example, Hart [13] describes the interval mutation operator, which replaces the value at a dimension of an individual with a uniformly selected value over the domain of that dimension. Similarly, Hart [13] describes a Cauchy mutation operator which adds the value of a Cauchy random variable to a dimension of an ....

....dimension of an individual with the value of a random variable, called replacement mutation. For example, Hart [13] describes the interval mutation operator, which replaces the value at a dimension of an individual with a uniformly selected value over the domain of that dimension. Similarly, Hart [13] describes a Cauchy mutation operator which adds the value of a Cauchy random variable to a dimension of an individual. Again, I distinguish between discrete mutation operators, which can generate a finite number of possible solutions a given solution, and indiscrete mutation operators, which can ....

[Article contains additional citation context not shown here]

W. E. Hart, Adaptive Global Optimization with Local Search, PhD thesis, University of California, San Diego, May 1994.

....of Autodock with simulated annealing [3, 9, 15, 14] and it is a good testbed for comparison with the genetic algorithm because of the effort that has gone into optimizing simulated annealing parameters. A preliminary comparison of genetic algorithms and simulated annealing in Autodock appears in [5]. 2 Global and Local Search Methods Docking is a difficult optimization problem, and successful search requires efficient local search of each attractor basin, as well as effective global sampling across the entire range of possible docking orientations. Earlier versions of Autodock relied ....

....the GA and local search operate on the same representation, so this mapping is trivial. Our previous work [6] leads us to use Lamarckian GA LS hybrids for the experiments presented here. In a GA LS hybrid, mutation plays a somewhat different role than it does in a GA without explicit local search [5]. Without an explicit local search operator, it must be the mutation operator that makes small, refining moves that are not efficiently made using crossover and selection alone. With an explicit local search operator, however, the local refinement requirement becomes unnecessary. Mutation is still ....

[Article contains additional citation context not shown here]

Hart, W.E. (1994). Adaptive Global Optimization with Local Search. PhD thesis, Computer Science & Engineering Department - University of California, San Diego. ftp://ftp.cs.sandia.gov/pub/papers/ wehart/thesis.ps.gz

....with simulated annealing (Morris et al. 1996) and it is a good testbed for comparison with the genetic algorithm because of the effort that has gone into optimizing simulated annealing parameters. A preliminary comparison of genetic algorithms and simulated annealing in Autodock appears in (Hart 1994). 2 GLOBAL AND LOCAL SEARCH Docking is a difficult optimization problem, and successful search requires efficient local search of each attractor basin, as well as effective global sampling across the entire range of possible docking orientations. Global local search hybrids may perform better on ....

....optimization problem, and successful search requires efficient local search of each attractor basin, as well as effective global sampling across the entire range of possible docking orientations. Global local search hybrids may perform better on optimization problems than either type separately (Hart 1994; Hart et al. 1994) Earlier versions of Autodock relied exclusively on an optimized variant of simulated annealing. Simulated annealing can be viewed as including both global and local search aspects, depending on whether it rejects a (locally) inferior alternative or jumps (globally) to a new, ....

[Article contains additional citation context not shown here]

Hart, W.E. (1994). Adaptive Global Optimization with Local Search. PhD thesis, CSE Dept. - University of California, San Diego.

....the term mutation for completely random, blind modifications. We can imagine more general definitions of local search and mutation which use information about a GA s current or previous populations. For example, the definition of local search does not encompass the methods described in Hart [7] that use statistics from the population to selectively apply local search. Our notion of mutation does not include the dynamically adjusted mutation operator used in evolutionary strategies, in which the standard deviation is adapted using the frequency of previously successful mutations. ....

....the bond lengths to be one. Further, it suggests that local search may not be as important when using this genotypic space since the solutions are already close to the nearby minima. We measured the utility of non Lamarckian local search for this problem by varying the frequency of local search [7, 8]. The experiment compared GAs using the following two genotypic spaces: a) the bond angles and bond lengths and (b) the bond angles. The space of atom coordinates was the phenotypic space for both GAs. A GA with floating point representation was used to search these genotypic spaces [7] Local ....

[Article contains additional citation context not shown here]

William E. Hart. Adaptive Global Optimization with Local Search. PhD thesis, University of California, San Diego, May 1994. The Role of Development in Genetic Algorithms 17

....B illustrate the toroidal nature of neighborhoods on A B Figure 3. A two dimensional grid used by GSGAs to define population subsets. this grid. We call a neighborhood like that shown in Figure 3 a NEWS neighborhood, since it only uses neighbors to the North, East, West and South. Recently, Hart [8] has described a coarse grain design for parallel GSGAs. Like SIMD GSGAs, this parallel GA uses a two dimensional toroidal grid that is distributed across the set of available processors. Thus each processor typically processes a set of solutions that is located on a subgrid of the entire grid. ....

....GSGAs will probably have a higher utilization of the parallel hardware, but processors may frequently be using border regions that are inconsistent with the state of the neighboring processors. 3. 2 Experimental Design The GSGAs used in our experiments are similar to the those described by Hart [8]. Mutation was performed by adding a value from a normally distributed random variable to a coordinate of a solution. We used a minimal neighborhood that included the two immediate neighbors along each dimension of the population grid (see Figure 3) The crossover rate was 0.8 and the mutation ....

[Article contains additional citation context not shown here]

W. E. Hart. AdaptiveGlobal Optimization with Local Search. PhD thesis, University of California, San Diego, May 1994.

No context found.

Hart, W. E. (1994). Adaptive global optimization with local search. Doctoral dissertation, University of California, San Diego, San Diego, CA.

No context found.

W. E. Hart. Adaptive global optimization with local search. Ph.D. Thesis, University of California, San Diego, 1994.

No context found.

W. E. Hart, "Adaptive global optimization with local search," Ph.D. Thesis, University of California, San Diego, 1994.

No context found.

W. E. Hart, 1994, "Adaptive Global Optimization with Local Search", Ph.D. Thesis, University of California, San Diego, May.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC