Generalized Clustering for Problem Localization

Abstract

functions. The initial standard cost was 600 and the unit, 60. In 7 s, shrink obtained a new cover of cost 455 and 52, respectively. Pseudo on the other hand in 1800 s obtained a solution of costs 388 and 48, respectively. In the same time exact obtained a standard-cost solution of 425 (with 69 cubes). For both exact and pseudo the time to generate the prime cubes (implicants) was 41 s. In neither case was a minimum obtained. JPL Problem P is concerned with control logic for Solar Energy Propulsion. It has 9 inputs and 3 outputs. The initial cover had 100 CARE cubes and 387 DON'TCARE cubes. Shrink was run 278 s to completion. Pseudo and exact were run for 2200 seconds. From an initial cost of 3087 the three routines obtained covers of reduced costs 2097, 1754, 1751, respectively. In the application, the cost reduction for the more expensive routines was justified. In most applications, e.g., an EAM transistorization project, the minimization procedures were nQt run to completion but the best solution so far obtained was used.

DOI: 10.1109/TC.1978.1675056

5 Figures and Tables

Cite this paper

@article{Fukunaga1978GeneralizedCF, title={Generalized Clustering for Problem Localization}, author={Keinosuke Fukunaga and Robert D. Short}, journal={IEEE Trans. Computers}, year={1978}, volume={27}, pages={176-181} }