Edit

Section 2.4  Other Metaheuristics

Simulated Annealing

Inspired by the physical annealing process of solids (crystals) - first melting then cooling very slowly, to obtain a minimum energy state.

Randomly generate neighbor solution and accept worse solution with reducing probabilities.

Tabu Search

Use short-term memory is the search process to avoid cycling through visited solutions.

Guided Local Search

Modify the evaluation function to escape from local optima

Iterated Local Search

When stuck at a local optima, perturb the solution to a new random solution.

Greedy Randomized Adaptive Search Procedures (GRASP)

Greedily construct initial solutions with randomness and start local searches from these initial solution.

Evolutionary Computation

Includes: Genetic Algorithm / Simulated Evolution /Stochastic Evolution

An example: http://genetic.moonlander.googlepages.com/home

Scatter Search

Keep a small population of reference solutions and combine them to create new solutions.

Stochastic Diffusion Search

Example: find best restaurant for all group members

Harmony Search

Mimicking the improvisation process of musicians

Cross-Entropy Method

Finding neighbor solution involves minimizing the cross-entropy or Kullback-Leibler divergence.

Particle Swarm Optimization

Individuals (particles) exchange information about local optimal among neighborhood and conduct search based on this information.

Extremal Optimization

Evolves a single solution and makes local modifications to the worst components. 

Common Sub-routines for Metaheuristics

Comparison of Metaheuristics 

Metaheuristics Type Solution Population Memory Utilization
Ant Colony Optimization constructive multiple Yes
Cross-Entropy Method local search single Yes
Evolutionary Computation local search multiple No
Extremal Optimization local search single No
Guided Local Search local search single Yes
Greedy Randomized Adaptive Search Procedures constructive single No
Harmony Search local search multiple Yes
Iterated Local Search local search single No
Particle Swarm Optimization local search multiple No
Simulated Annealing local search single No
Stochastic Diffusion Search constructive multiple No
Scatter Search local search single No
Tabu Search local search single Yes

Pros and Cons of Metaheuristics

 

P vs NP