Hagen, L. W., & Kahng, A. B. (1997). Combining problem reduction and adaptive multistart: A new technique for superior iterative partitioning. IEEE Transactions on CAD, 16 (7), 709--717.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....for a good start state. Many other sensible heuristics for adaptive restarting have been shown e ective in the literature. The widely applied methodology of tabu search (Glover Laguna, 1993) is fundamentally a set of adaptive heuristics for escaping local optima, like CLO s kick steps. Hagen and Kahng s (1997) Clustered Adaptive Multi Start achieves excellent results on the VLSI netlist partitioning task; like CLO, it alternates between search with high level operators (constructed adaptively by clustering elements of previous good solutions) and ordinary local search. Jagota s Stochastic Steep ....

Hagen, L. W., & Kahng, A. B. (1997). Combining problem reduction and adaptive multistart: A new technique for superior iterative partitioning. IEEE Transactions on CAD, 16 (7), 709-717.

....of problem size reduction provides another way to avoid the central limit catastrophe. It uses clustering to dramatically reduce the size of the solution space. This approach, combined with multi start, has been particularly successful for circuit partitioning in VLSI computer circuit design [41] [50] [52] Full Cooperation Cooperation with Penalties Communication Only Independent Agents Single Agent Sequential Multi Start Figure 2: Hierarchy of dominance between di erent models of adaptive annealing. Each arrow points from a dominating to a dominated model. Neural Computing Surveys ....

L. W. Hagen and A. B. Kahng, \Combining Problem Reduction and Adaptive Multi-Start: A New Technique for Superior Iterative Partitioning", IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 16 (7), 709-17 (1997).

....similar solutions. Their abstract summarizes three algorithms that make the genetic algorithm s modeling function explicit, consequently improving optimization performance. Buntine, Su and Newton (x ) learn a generative model in the problem of hyper graph partitioning, crucial in VLSI design [3]. The model is in the form of a clustering of the graph nodes, based on a statistical analysis of the best solutions found so far in the search. The clustering e ectively scales down the size of the search space, enabling good new candidate solutions to be generated very quickly. Neural Computing ....

L. W. Hagen and A. B. Kahng. Combining problem reduction and adaptive multi-start: A new technique for superior iterative partitioning. IEEE Transactions on CAD, 16(7):709-717, 1997.

....literature. The widely applied methodology of tabu search (Glover Laguna, 1993) is fundamentally a set of adaptive heuristics for escaping local optima, like CLO s kick steps. Hagen and Kahng s Clustered Adaptive Multi Start achieves excellent results on the VLSI netlist partitioning task (Hagen Kahng, 1997); like CLO, it alternates between search with high level operators (constructed adaptively by clustering elements of previous good solutions) and ordinary local search. Jagota s Stochastic Steep Descent with Reinforcement Learning heuristically rewards good starting states and punishes poor ....

Hagen, L. W., & Kahng, A. B. (1997). Combining problem reduction and adaptive multistart: A new technique for superior iterative partitioning. IEEE Transactions on CAD, 16 (7), 709--717.

....and list some of the cost functions used in the prior research to build good clusters. 2.3. 1 Using Clustering to Reduce Problem Complexity Whether the problem is partitioning or placement, the virtues of using bottom up netlist clustering are well documented in [Sun95] Roy93] Shin93] [Hage97] [Kary97] Alpe97c] The primary goal of this clustering is to reduce the problem size so that a smaller and more easily solvable problem is obtained. This assists in decreasing the time required for iterative algorithms to obtain a good solution for the overall problem. A clustering groups ....

....algorithms whose performance tend to degrade as the problem size and complexity increase. Hagen and Kahng suggest that the advantage offered by clustering in reducing the problem size permits the algorithm operating on the condensed problem to focus on the most difficult and time consuming portion [Hage97]. Both [Sun95] and [Roy93] state that through effective netlist clustering, the number of clusters to manipulate, the number of inter cluster nets and pins, and the average fanout of the remaining nets are all substantially reduced. This can decrease the computation time required by an order of ....

L. W. Hagen and A. B. Kahng, "Combining Problem Reduction and Adaptive Multistart: A New Technique for Superior Iterative Partitioning," IEEE Transactions on Computer-Aided Design, vol. 16, no. 7, July 1997, pp. 709-717.

....3 Opt 0.66 44 0.54 32 Lin Kernighan 0.73 54 0.57 34 LSMC (3 Opt) 0.69 41 0.40 19 Table 2: Correlations between distance and cost for the five heuristics applied to ATT532. Based on the unique minima resulting from 2,500 runs of each heuristic. between cost and distance. However, Hagen and Kahng [5] have shown for circuit partitioning at least, that this relationship deteriorates for lower cost solutions (i.e. those produced by more powerful heuristics such as Fiduccia Mattheyses [4] In other problem formulations we also find weaker cost distance relationships than in the TSP, although in ....

....the TSP that constrain edges in later descents if they are common to all of the best tours in earlier descents. This strategy is very similar to a multi start approach suggested by Lin and Kernighan in their 1973 paper, except that we now freeze edges common to the best previous solutions (cf. [5]) rather than only the edges common to all previous solutions. ....

L. Hagen and A. B. Kahng, "Combining Problem Reduction and Adaptive Multi-Start: A New Technique For Superior Iterative Partitioning", to appear in IEEE Trans. Computer Aided Design, 1995.

....The most widely used iterative algorithm is that of Fiduccia and Mattheyses (FM) 5] Wei and Cheng [16] use an adaptation of [5] to address the ratio cut objective. The main weakness of FM and its variants is that solution quality is not stable , i.e. it is not predictable. Hagen and Kahng [7] report that the distribution of solution costs (nets cut) for independent random FM executions on ACM SIGDA benchmark netlists is approximately normal. This implies not only that the average FM local minimum is significantly worse than the best possible FM solution, but also that FM must be run ....

....kick move yielded the best performance, followed by the net removal kick move. 3) In general, the larger values of MoveSize resulted in better performance; the randomized MoveSize performed well overall. We also compare LSMC with the best previous partitioning results in the literature (e.g. [1, 4, 7]) Table 2 and Table 3 show the comparison using unit are and actual area respectively. Each entry is based on 50 runs of LSMC, where 3 For the Biomed benchmark in Table 1, our results are based on 500 passes of FM. Recall that an FM pass is linear time; it moves and locks each node exactly ....

[Article contains additional citation context not shown here]

L. Hagen and A.B. Kahng, "Combining Problem Reduction and Adaptive Multi-Start: a New Technique for Superior Iterative Partitioning", to appear in IEEE Trans. on Computer-Aided Design.

No context found.

Hagen, L. W., & Kahng, A. B. (1997). Combining problem reduction and adaptive multistart: A new technique for superior iterative partitioning. IEEE Transactions on CAD, 16 (7), 709--717.

No context found.

L. Hagen and A. B. Kahng. Combining problem reduction and adaptive multistart: A new technique for superior iterative partitioning. IEEE Transactions on Computer Aided Design, 1995. (to appear).

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