Academia.eduAcademia.edu

Figure from:General iterative heuristics for VLSI multiobjective partitioning

See full PDFdownloadDownload figure
Table 1: Comparison between costs of the best solutions generated by GA and TS to its directed search approach and its higher greediness tendency as compared with GA to obtain a good solution.

Table 1 Comparison between costs of the best solutions generated by GA and TS to its directed search approach and its higher greediness tendency as compared with GA to obtain a good solution.

Related Figures (4)

Figure 2: Comparison between GA and TS for the circuit $13207 with respect to execution time in seconds. The results shown are the best case results obtained after the tuning of various algorithmic parameters of GA and TS (only one time for all circuits). In the case of GA the population size is 10, the crossover used is simple with a probability equal to 0.99, while for mutation it is 0.01. In case of TS, the size of neighborhood is also 10, while Tabu list size is chosen to be 0.1 the size of the cir- cuit. From the results, it is clear that TS performed better than GA for most of the circuits in terms of the quality of the best solution as well as run time. In terms of quality of solution, TS consistently performs better, and the advantage of TS over GA gets emphasized when the size of the circuit gets bigger. Also execution time of GA increases significantly with the increase in circuit complexity. The higher execution time of GA can be attributed to its parallel nature i.e., a population of solutions is to be processed in each generation. Fig. 2 shows the performance of TS and GA against execution time in seconds. Itis clearly noticed that TS is by far faster and of better final quality. Fig. 3 and Fig. 4 show the trend of solution’s (a) cut- set, (b) delay, (c) power, (d) balance, (e) average fitness, (f) best fitness for GA and TS respectively, in case of circuit $12307. It is clear from the shown plots that TS achieves a membership that is better than that reached by GA.
Figure 2: Comparison between GA and TS for the circuit $13207 with respect to execution time in seconds. The results shown are the best case results obtained after the tuning of various algorithmic parameters of GA and TS (only one time for all circuits). In the case of GA the population size is 10, the crossover used is simple with a probability equal to 0.99, while for mutation it is 0.01. In case of TS, the size of neighborhood is also 10, while Tabu list size is chosen to be 0.1 the size of the cir- cuit. From the results, it is clear that TS performed better than GA for most of the circuits in terms of the quality of the best solution as well as run time. In terms of quality of solution, TS consistently performs better, and the advantage of TS over GA gets emphasized when the size of the circuit gets bigger. Also execution time of GA increases significantly with the increase in circuit complexity. The higher execution time of GA can be attributed to its parallel nature i.e., a population of solutions is to be processed in each generation. Fig. 2 shows the performance of TS and GA against execution time in seconds. Itis clearly noticed that TS is by far faster and of better final quality. Fig. 3 and Fig. 4 show the trend of solution’s (a) cut- set, (b) delay, (c) power, (d) balance, (e) average fitness, (f) best fitness for GA and TS respectively, in case of circuit $12307. It is clear from the shown plots that TS achieves a membership that is better than that reached by GA.
Figure 3: Performance of Ga for the circuit s13207.
Figure 3: Performance of Ga for the circuit s13207.
Figure 4: Performance of TS for the circuit s13207.
Figure 4: Performance of TS for the circuit s13207.