24vls_neighborhoodsearch

24vls_neighborhoodsearch - Very Large Scale Neighborhood...

Info iconThis preview shows pages 1–12. Sign up to view the full content.

View Full Document Right Arrow Icon
Very Large Scale Neighborhood Search Collaborators include: Ravi Ahuja, Ozlem Ergun, Abraham Punnen, Dushyant Sharma
Background image of page 1

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

View Full DocumentRight Arrow Icon
Neighborhood Search Combinatorial Optimization : minimize (f(S) : S ¡# F) ¡ f is typically linear, ¡ F is finite Neighborhood Function : ¡ For each S ¡# F, there is a neighborhood N(S); ¡ We say that S is a local optimum if f(S) d# f(T) for all T ¡# N(S); 2
Background image of page 2
Neighborhood Search Neighborhood Search (local improvement algorithm) begin initialize with some S ¡# F; while S is not a local optimum do replace S by some T ¡# N(S) such that f(T) < f(S); end 3
Background image of page 3

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

View Full DocumentRight Arrow Icon
4 TSP and 2-exchanges 1 2 3 4 5 6 7 8 9 10 The original tour 1 2 3 9 10 4 5 6 7 8 A 2-neighbor of the original tour We say that a tour T’ is a 2- neighbor of a tour T if it is possible to obtain T’ from T by adding two edges and deleting two edges. The operation is called a 2-exchange. T’ = T + (3,8) + (4,9) - (3,4) - (8,9). Obtained by the operation Flip[4,8] When we say city i, we really mean the city that is in position i of the current tour. (Or you may assume that the current tour is 1, 2, 3, . .., n)
Background image of page 4
5 A non-optimal tour 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Background image of page 5

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

View Full DocumentRight Arrow Icon
2 3 4 5 6 7 6 1 13 14 15 16 17 18 19 20 Improving using a two exchange 8 9 10 11 12 7 6 5 4 3 2 A 2-exchange is obtained by deleting two edges of the tour and adding two edges
Background image of page 6
Neighborhood Search We say that a tour T’ is a 2-neighbor of a tour T if it is possible to obtain T’ from T by adding two edges and deleting two edges. This can be generalized to 3-neighbors , and k-neighbors . Neighborhood Search: begin start with some tour T. while there is a neighbor T’ of T with a lower objective value, replace T by T’ end 7
Background image of page 7

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

View Full DocumentRight Arrow Icon
More on Neighborhood Search A solution is 2-optimum if there is no 2-neighbor that has a lower length. A solution is locally optimum if there is no neighbor with an improved objective. Neighborhood search in its “vanilla” form stops when a locally optimum solution is found. Most neighborhood search relies on neighborhoods that can be searched exhaustively. 8
Background image of page 8
Some simple extensions Extension 1. For k = 1 to 1000, select a random tour T. Find a locally optimal tour starting with T. Then choose the best of these tours. Extension 2. 1. Let T be a tour 2. Find a locally optimal tour T’ starting at T 3. perturb T’ (perhaps using a random 5 exchange) 4. replace T by T’ and return to step 1. Other extensions: simulated annealing and tabu search 9
Background image of page 9

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

View Full DocumentRight Arrow Icon
Very Large Scale Nbhd (VLSN) search This talk: Focuses on VERY LARGE neighborhoods that can be searched very efficiently (preferably in polynomial time) or are searched heuristically. often exponentially large neighborhoods Rule of Thumb for Larger Neighborhoods: improved local optima greater search time 10
Background image of page 10
relation to other search techniques min cost flows, cycle canceling Multicommodity flows: column generation But both these problems are special case of linear programming. We will use this technique for hard problems as well.
Background image of page 11

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

View Full DocumentRight Arrow Icon
Image of page 12
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 03/15/2010 for the course IE 505 taught by Professor Yok during the Spring '10 term at Galatasaray Üniversitesi.

Page1 / 53

24vls_neighborhoodsearch - Very Large Scale Neighborhood...

This preview shows document pages 1 - 12. Sign up to view the full document.

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