24vls_neighborhoodsearch

# 24vls_neighborhoodsearch - Very Large Scale Neighborhood...

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

Very Large Scale Neighborhood Search Collaborators include: Ravi Ahuja, Ozlem Ergun, Abraham Punnen, Dushyant Sharma

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

View Full Document
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
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

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

View Full Document
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)
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

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

View Full Document
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
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

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

View Full Document
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
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

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

View Full Document
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
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.

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

View Full Document
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
Ask a homework question - tutors are online