Natural Optimization Algorithms

1
Natural Optimization Algorithms

• Chapter 10

2
Practical Swarm Optimization

• start with a random population (similar to GAs)
• continuous encoding (i.e., not binary)
• each particle (i.e., gene in GA terminology)
  moves in the search space
• pinew pioldvinew
• vinewviold r1 a1 (plocal - piold) r2 a2
  (pglobal - piold))
• a1a2 2
• r1, r2 - random numbers
• plocal - best cost in present population
• pglobal - best cost found so far

3
(No Transcript)
4
(No Transcript)
5
Genetic Programming

• programs are trees
• e.g. represents
  (xx)x
• x
• x x
• suppose we want to find a program that
  interpolates sin(x) for x in 0,?
• the cost function will be the difference between
  the found points and the actual points

6
Genetic Programming

• crossover choose two nodes at random, exchange
  the corresponding sub-tress
• mutation
• substitute a randomly selected sub-tree with a
  randomly generated sub-tree
• replace the function or variable in a given node
• note that leaf nodes are variables/constants and
  non-leaf nodes are functions

7
Applications

• not a single best solution
• the cost function may evolve
• design of circuits, antennas, dynamically
  changing hardware

8
(No Transcript)
9
Future of GAs

• Complex problems that consists of a bunch of
  sub-problems
• e.g. airplane scheduling problem
• schedule flights, plane routing, crew,
  passengers, luggage
• optimizing each problem may be easy, but those
  problems are not independent
• application with GPS, e.g. how to get to a point,
  car navigation systems