Lecture 6 - Improving Local Search
Lecture 6 - Improving Local Search
Lecture 6
Heuristic Optimization Methods:
Improving Local Search
Slides prepared by Lea Skorin-Kapov
Zavod za telekomunikacije 3
Outline
University of Zagreb, Croatia
Zavod za telekomunikacije 4
Multistart local search
University of Zagreb, Croatia
Already
found!
Still not
1st found local
2ndfound local The best seen necessarily
optimum solution is the optimum globally
final one optimal…
Zavod za telekomunikacije 5
Multistart local search
University of Zagreb, Croatia
Zavod za telekomunikacije 6
Iterated local search (ILS)
University of Zagreb, Croatia
Zavod za telekomunikacije 7
Iterated local search (ILS)
University of Zagreb, Croatia
objective
initial solution
LS
𝑠0 perturbed solution
𝑠∗
𝑠′
first local optimum
𝑠∗′
second local optimum
search space
If S* is the set of all locally optimal solutions, we want to explore S* using a walk from
one local optimum to another. If 𝑠∗′ passes an acceptance test, it becomes the next
element in the walk. Otherwise go back to 𝑠∗
Zavod za telekomunikacije 8
Iterated local search (ILS)
University of Zagreb, Croatia
Zavod za telekomunikacije 9
Iterated local search (ILS)
University of Zagreb, Croatia
Design choices:
1. Local search: any single-solution metaheuristic
can be used (e.g., simple descent, tabu search,
SA…)
2. Perturbation method: the perturbation operator
may be seen as a large random move of the
current solution.
Goal: keep some part of the solution and perturb strongly
another part of the solution
3. Acceptance criteria: defines the conditions the
new local optima must satisfy to replace the current
solution.
Zavod za telekomunikacije 10
ILS: perturbation method
University of Zagreb, Croatia
Zavod za telekomunikacije 11
ILS: acceptance criteria
University of Zagreb, Croatia
weak strong
selection selection
Diversification Intensification
Zavod za telekomunikacije 13
Outline
University of Zagreb, Croatia
Zavod za telekomunikacije 14
Variable Neigborhood Search (VNS)
University of Zagreb, Croatia
Key concepts:
Fact 1: a local minimum w.r.t. one neighborhood
structure is not necessarily so for another
Fact 2: A global minimum is a local minimum w.r.t. all
possible neighborhood structures
Fact 3: For many problems, local minima w.r.t. one or
several Nk (neighborhood structures) are relatively
close to one another.
Zavod za telekomunikacije 15
Variable Neigborhood Search (VNS)
University of Zagreb, Croatia
Basic idea:
successively explore a set of predefined neighborhoods to provide
a better solution
Systematically change neighborhood both in a descent phase to
find a local optimum and in a perturbation phase to get out of a
correspoding valley
Zavod za telekomunikacije 16
Variable Neigborhood Descent (VND)
University of Zagreb, Croatia
Improving neighbor
Non-improving neighbor
Zavod za telekomunikacije 17
Variable Neigborhood Descent (VND)
University of Zagreb, Croatia
w.r.t. the order in which neighborhoods are applied: the most popular
strategy is to rank the neighborhoods following the increasing order of their
complexity (e.g., the size of the neighborhoods)
Zavod za telekomunikacije 18
Basic VNS
University of Zagreb, Croatia
Zavod za telekomunikacije 19
Basic VNS
University of Zagreb, Croatia
Zavod za telekomunikacije 22
Guided Local Search (GLS)
University of Zagreb, Croatia
Basic idea:
dynamically change objective function according to the
already generated local optima
features of the obtained local optima are used to
transform the objective function
allows modification of the landscape structure to escape
from the obtained local optima
Zavod za telekomunikacije 23
GLS
University of Zagreb, Croatia
objective
objective penalization
𝑠0
Zavod za telekomunikacije 24
GLS
University of Zagreb, Croatia
Zavod za telekomunikacije 25
GLS
University of Zagreb, Croatia
Zavod za telekomunikacije 27
GLS
University of Zagreb, Croatia
penalties
cost (objective)
function augmented cost
function
source: Voudouris, Christos & Tsang, Edward & Alsheddy, Abdullah. (2010). Guided Local Search. 10.1007/978-1-4419-1665-5_11.
Zavod za telekomunikacije 28