I. Hong, A. B. Kahng, and B. R. Moon. Improved large-step Markov chain variants for the symmetric TSP. Journal of Heuristics, 3(1):63--81, 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....limited interaction between the population elements. Such extensions have strong similarities to well known population based search metaphors such as evolutionary algorithms [18, 19] and, in particular, memetic algorithms [6, 20] Population based extensions of ILS were independently proposed in [21, 22] and applied to the TSP and the QAP. In particular, Stutzle reported results for different levels of interaction among the individual ILS search threads: no interaction, replace worst and population ILS. No interaction, as the name implies, is simply a population of ILS trajectories run in ....

....new solution is inserted into the population and the process is iterated. Stutzle reported varying degrees of success with this approach. The performance of all three variants was found to be similar and appeared to depend strongly on the given problem instance. Similarly, Hong, Kahng, and Moon [21] presented computational results that did not indicate a significant improvement of the population based ILS over a single trajectory ILS for the TSP. 4 Our ILS Algorithms for the MPTTP The MPTTP problem intuitively involves two subproblems, the problem of partitioning the set of jobs, where ....

Hong, I., Kahng, A.B., Moon, B.R.: Improved large-step Markov chain variants for the symmetric TSP. Journal of Heuristics 3 (1997) 63--81

....annealing component into the heuristic. Results for five descendants of this approach were submitted to the Challenge, differing both in the underlying local search heuristic used and in the methods used for generating the double bridge kicks. Variants using more elaborate kicks have been studied [20, 41], but none have yet proved competitive on large instances with the best of the heuristics presented here. Together, these five include all the current top performers known to the authors for this region of the time quality tradeoff space. None of the implementations use simulated annealing, and ....

I. Hong, A. B. Kahng, and B. Moon. Improved large-step Markov chain variants for the symmetric TSP. J. Heuristics, 3:63--81, 1997.

....0.13 0.0011 1.23 0.0031 0.0072 0.033 0.062 2. VND [16] leads to a very powerful local search for the SMTWTP. 3. For the SMTWTP the optimization of the ILS algorithm leads to improved performance. Yet, the improvement due to these optimizations is not as spectacular as observed for other problems [6, 19, 21], possibly due to the powerful local search. 4. The SMTWTP instances from ORLIB do not pose a challenge for state of the art algorithms. Clearly, these conclusions do raise further questions. The optimization of ILS for the SMTWTP was certainly rather straightforward and heuristic. We conjecture ....

I. Hong, A. B. Kahng, and B. R. Moon. Improved large-step Markov chain variants for the symmetric TSP. Journal of Heuristics, 3(1):63--81, 1997.

....far better than random restart. Thus one should keep in mind that the optimization of an iterated local search may require more than the optimization of the individual components. 3.2. 2 Adaptive perturbations The behavior of ILS for the QAP and also for other combinatorial optimization problems [35, 60] shows that there is no a priori single best size for the perturbation. This motivates the possibility of modifying the perturbation strength and adapting it during the run. One possibility to do so is to exploit the search history. For the development of such schemes, inspiration can be taken ....

....is an appropriate way to achieve search space diversification. Thus population based methods are desirable if their complexity is not overwhelming. Because of this, population based extensions of ILS are promising approaches. To date, several population based extensions of ILS have been proposed [1, 35, 61]. The approaches proposed in [35, 61] keep the simplicity of ILS algorithms by maintaining unchanged the perturbations: one parent is perturbed to give one child. But given that there is a population, the evolution depends on competition among its members and only the fittests survive. One can ....

[Article contains additional citation context not shown here]

I. Hong, A. B. Kahng, and B. R. Moon. Improved large-step Markov chain variants for the symmetric TSP. Journal of Heuristics, 3(1):63--81, 1997.

.... boost over the original Lin Kernighan scheme, as demonstrated in the computational study of Johnson [1990] Further results comparing the two approaches can be found in Reinelt [1994] Junger, Reinelt, and Rinaldi [1995] Codenotti, Manzini, Margara, and Resta [1996] Johnson and McGeoch [1997] Hong, Kahng, and Moon [1997], and Neto [1999] These papers offer very good implementations for problem instances ranging in size from several hundred cities up to several thousand cities. In this note, we discuss a number of issues involved in obtaining efficient implementations of Chained Lin Kernighan for instances having ....

....that for each of our local kicks, the wider variants are preferable for long runs, while the intermediate width kicks are preferable for short and medium runs. Our reported results are restricted to double bridge kicks, but there is no strong argument favoring these over other kicking structures. Hong, Kahng, and Moon [1997] tested the use of k exchange kicks for k varying from 2 up to 50; they report that several values of k work well on their 318 city, 532 city, and 800 city test instances. A quite different kick Table 3: Kicking Strategy Kick Short Run Medium Run Long Run random 0.749 0.398 0.204 random, ....

I. Hong, A. B. Kahng, and B. Moon (1997). "Improved large-step markov chain variants for the symmetric TSP", Journal of Heuristics 3, 63--81.

.... originated with Baum in 1986 [2] also in the context of the TSP; note that the particular choice of kick move can strongly influence the perceived utility of LSMC, as can be seen from the contrasting results of [2] and [10] as well as the more detailed kick move studies of LSMC for the TSP in [9]. Figure 1 describes the LSMC approach, following the presentation in [13] but adapted to the partitioning domain, i.e. the greedy local optimization procedure is assumed to be the FM algorithm. 1 As can be seen from the Figure, LSMC actually performs simulated annealing over the set of FM ....

....f generate a random number t 2 [0; 1) if ( t exp( Gamma diff =temp i ) P i 1 = P i ; else P i 1 = P i ; g 2.4. if ( cost(Pbest) cost(P i ) Pbest = P i ; 3. return Pbest ; Figure 1. LSMC algorithm template. on variants of zero temperature LSMC, we note that Hong et al. [9] have recently studied the utility of non zero temperatures in applying LSMC to the TSP, and have shown that significant performance improvements can result. In work related to our present application, 13] briefly describes the application of LSMC to graph partitioning on random geometric graphs, ....

[Article contains additional citation context not shown here]

I. Hong, A. B. Kahng and B.-R. Moon, "Improved LargeStep Markov Chain Variants for the Symmetric TSP", UCLA Computer Science Dept. Tech. Report UCLA-CSD950035, 1995.

No context found.

I. Hong, A. B. Kahng, and B. R. Moon. Improved large-step Markov chain variants for the symmetric TSP. Journal of Heuristics, 3(1):63--81, 1997.

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