文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > Simulated Annealing for Graph Bisection
侵权投诉

Simulated Annealing for Graph Bisection

Simulated Annealing for Graph Bisection

Note that a typical random partition has k= O(n1=2), while the canonical partiti

on (which with high probability is the minimum) is uniquely characterized by k= n. As we will see, the cost of a state is equal to a function of k plus an error term which is small (with high probability). Thus the sequence k(t) of imbalances after t attempted moves can be used to study the progress of annealing. That is, while the transition probabilities k 7! k? 1, k 7! k, and k 7! k+1 depend on the full state, under appropriate conditions they can be approximated from k alone, making k(t) a tractable\near Markov chain". Throughout this paper the phrase with overwhelming probability will mean\with probability 1? O( exp(?nc)) for some constant c> 0". The full probability space under consideration is the cross product of the space of random graphs, the random proposed moves over all steps, and the random coin tosses at all steps. We will write EG;M to denote the expectation of some quantity over random graphs and moves, PrM to denote a probability over random moves alone, and so forth. Finally, in Sections 4 and 5 it is presumed that the sequence of all proposed moves is speci ed in advance: this must be done in such a way that the acceptance or rejection of a previous move does not jeopardize the legality or uniform randomness of later moves. The following means of specifying moves satis es this, and is a convenient concrete model for implementation of the algorithm. Consider the collections L and R to be ordered lists of vertices rather than mere sets, so that a state= (L; R) is a permutation. (Conversely, for a state speci ed by the partition, L and R will refer to the rst and last 2n vertices.) Let the move (i; j ) indicate the exchange of Li and Rj: the i'th vertex in L and the j 'th vertex in R. In annealing, moves will be generated by choosing i and j uniformly at random from f1::: 2ng.

2.3 Randomness and move generation

3 Typical Transition Probabilities The transition k 7! k+ 1 occurs i a move taking k to k+ 1 is both proposed and accepted.

Claim 3.1 For any graph G and any partition with imbalance k, the probability that a proposed move would increase k is

1 k k2 (1) 4? 2n+ 4n2: Proof A proposed move (i; j ) increases k if Li is white, and Rj is black, which has probability ((n? k)=2n)2 . For a given k-bisection, and k-increasing move M, let X+= X+ (G) be the random variable representing the cost increase that occurs when move M is applied to the partitioning of the graph G. Recall that A(x)= (1+ exp(x=T ))?1 is the acceptance function. Note that A(X+ ) is the probability that a given k-increasing move is accepted. Pr propose k 7! k+ 1]= M 5

上一页第5页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email