Optimization by Building and Using Probabilistic Models
Organized by: Peter A.N. Bosman, Jörn Grahl, Kumara Sastry and Martin Pelikan.
Part of the Genetic and Evolutionary Computation Conference (GECCO-2006)
Seattle, WA. on July 08 - 12 (Saturday - Wednesday), 2006
Check the program section for slides from the workshop!
Workshop date: Sunday, July 9
Workshop topic:
Genetic- and evolutionary algorithms (GEAs) evolve a
population of candidate solutions to a given optimization
problem using two basic operators: (1) selection and (2)
variation. Selection introduces a pressure toward high-quality
solutions, whereas variation ensures exploration of the
space of all potential solutions. Two variation operators
are common in current genetic{ and evolutionary computation
(GEC): (1) crossover, and (2) mutation. Crossover creates
new candidate solutions by combining bits and pieces
of promising solutions, whereas mutation introduces slight
perturbations to promising solutions to explore their immediate
neighborhood. However, fixed, problem-independent
variation operators often fail to effectively exploit important
features of high-quality selected solutions. One way to make
variation operators more powerful and flexible is to replace
traditional variation of GEAs by the following two steps:
1. Build a probabilistic model of the selected promising
solutions, and
2. sample the built model to generate a new population
of candidate solutions.
Algorithms based on this principle are commonly called
estimation-of-distribution algorithms (EDAs) but are also
known as probabilistic model-building genetic algorithms
(PMBGAs) and as iterated density{estimation evolutionary
algorithms (IDEAs).
