Heavy-tailed distributions in combinatorial search

作者: B. Selman , C. P. Gomes , N. Crato

DOI:

关键词:

摘要: Combinatorial search methods often exhibit a large variability in performance. We study the cost profiles of combinatorial procedures. Our reveals some intriguing properties such profiles. The distributions are characterized by very long tails or heavy tails. will show that these best general class have no moments (i.e., an infinite mean, variance, etc.). Such non-standard recently been observed areas as diverse economics, statistical physics, and geophysics. They closely related to fractal phenomena, whose was introduced Mandelbrot. believe this is first finding purely computational setting. also how random restarts can effectively eliminate heavy-tailed behavior, thereby dramatically improving overall performance procedure.

参考文章(0)