Algorithms_Part16 - S. Dasgupta, C.H. Papadimitriou, and...

Info iconThis preview shows pages 1–4. Sign up to view the full content.

View Full Document Right Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: S. Dasgupta, C.H. Papadimitriou, and U.V. Vazirani 301 Figure 9.8 Local search. Figure 9.7 shows a specific example of local search at work. Figure 9.8 is a more abstract, stylized depiction of local search. The solutions crowd the unshaded area, and cost decreases when we move downward. Starting from an initial solution, the algorithm moves downhill until a local optimum is reached. In general, the search space might be riddled with local optima, and some of them may be of very poor quality. The hope is that with a judicious choice of neighborhood structure, most local optima will be reasonable. Whether this is the reality or merely misplaced faith, it is an empirical fact that local search algorithms are the top performers on a broad range of optimization problems. Let’s look at another such example. 9.3.2 Graph partitioning The problem of graph partitioning arises in a diversity of applications, from circuit layout to program analysis to image segmentation. We saw a special case of it, BALANCED CUT , in Chapter 8. G RAPH PARTITIONING Input: An undirected graph G = ( V, E ) with nonnegative edge weights; a real number α ∈ (0 , 1 / 2] . Output: A partition of the vertices into two groups A and B , each of size at least α | V | . Goal: Minimize the capacity of the cut ( A, B ) . 302 Algorithms Figure 9.9 An instance of GRAPH PARTITIONING , with the optimal partition for α = 1 / 2 . Vertices on one side of the cut are shaded. Figure 9.9 shows an example in which the graph has 16 nodes, all edge weights are or 1 , and the optimal solution has cost . Removing the restriction on the sizes of A and B would give the MINIMUM CUT problem, which we know to be efficiently solvable using flow techniques. The present variant, however, is NP-hard. In designing a local search algorithm, it will be a big convenience to focus on the special case α = 1 / 2 , in which A and B are forced to contain exactly half the vertices. The apparent loss of generality is purely cosmetic, as GRAPH PARTITIONING reduces to this particular case. We need to decide upon a neighborhood structure for our problem, and there is one obvious way to do this. Let ( A, B ) , with | A | = | B | , be a candidate solution; we will define its neighbors to be all solutions obtainable by swapping one pair of vertices across the cut, that is, all solutions of the form ( A-{ a } + { b } , B-{ b } + { a } ) where a ∈ A and b ∈ B . Here’s an example of a local move: We now have a reasonable local search procedure, and we could just stop here. But there is still a lot of room for improvement in terms of the quality of the solutions produced. The search space includes some local optima that are quite far from the global solution. Here’s one which has cost 2 . S. Dasgupta, C.H. Papadimitriou, and U.V. Vazirani 303 What can be done about such suboptimal solutions? We could expand the neighborhood size to allow two swaps at a time, but this particular bad instance would still stubbornly resist.to allow two swaps at a time, but this particular bad instance would still stubbornly resist....
View Full Document

Page1 / 20

Algorithms_Part16 - S. Dasgupta, C.H. Papadimitriou, and...

This preview shows document pages 1 - 4. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online