Probabilistic modelling in continuous metaheuristic optimization using mixture models
Speaker: Naveen Kumar, ITEE
When: 2006-06-14 11:00:00
Venue: 78-420
Host: Marcus Gallagher
Abstract:Metaheuristics have received considerable interest as a class of
algorithms that provide good performance on a range of difficult
optimization problems. Well-known examples of metaheuristics are
evolutionary algorithms (e.g genetic algorithms), simulated
annealing, ant colony optimization and tabu search. A recent
development in the area has been a class of population-based
optimization algorithms that use probabilistic modelling and machine
learning techniques to control and direct the search process. These
algorithms are most commonly known as Estimation of Distribution
Algorithms (EDAs).
The aim of this project is to develop more flexible and efficient
continuous EDAs using Gaussian mixture models trained using the
Expectation Maximization (EM) algorithm. This model has several
desirable properties and advantages over existing EDA modelling
techniques, such as the ability to model multimodal distributions
and the ability to capture dependency information implicitly,
without the need for complex modelling using probabilistic graphical
models. The performance of these algorithms will be studied and
evaluated. The influence of the user-parameters of the algorithms
will be investigated, with particular focus on the covariance
structure of Gaussian components in the mixture model, as this is
likely to be the most significant factor in controlling the
complexity (and therefore efficiency) of the model. The algorithms
will also be compared with existing EDAs and other EAs based on
simulations and results in the literature. It is expected that the
GMMEDA will provide a powerful, robust and efficient algorithm for
certain kinds of hard continuous optimization problems.
Biography:(biography unavailable)
Type: Ph.D confirmation
Contact:Marcus Gallagher, seminar host (marcusg@itee.uq.edu.au)
or Guido Governatori (ITEE seminar co-ordinator)
(guido@itee.uq.edu.au)