ITEE Ph.D confirmation seminar: Bo Yuan, 03.00PM, Wed 29 Oct 2003

A Statistical Modeling Approach to Evolutionary Algorithms and Their Empirical Evaluation

Speaker: Bo Yuan, ITEE

When: 03.00PM, Wednesday 29 Oct 2003

Venue: 78-622

Host: Dr. Marcus Gallagher

Abstract:

  There are three major issues to be pursued in my PhD work.

  1. How to effectively evaluate and test EAs?

  Driven by the success of EAs in solving many challenging problems,
  much effort has been devoted to designing new algorithms or trying
  to improve existing algorithms. As a contrast, little attention has
  been paid to a question:How to evaluate and compare those EAs?

  Due to their inherent dynamic and stochastic property, it is very
  difficult to theoretically analyze and describe the behavior of
  these algorithms. As a consequence, our current theory can only be
  applied to some idealized or simplified situations.

  Unfortunately,current methodology of doing empirical experiments is
  also far from perfect. For example, many of those benchmark problems
  widely used have proved to be inappropriate. As a result, experiment
  results established on those benchmark problems are questionable.

  Our objective is to build a principled framework for evaluating EAs
  using landscape generators and various statistical methods.

  2. Estimation of Distribution Algorithms

  Although EAs have a lot of advantages compared to other traditional
  optimization algorithms, those genetic operators (crossover/muation)
  are generally problem independent and cannot explicitly capture the
  structure of the problem to be solved. As a result, if there are
  some complex interactions among variables, EAs can be of very low
  efficiency.

  Recently, a new class of EAs based on statistical models has
  emerged.  These algorithms are called EDAs, which discard genetic
  operators and instead build an abstract model (e.g.,Bayesian
  networks) to capture the dependences among variables and use this
  model to gudie the searching process.

  Our work in this part will mainly focus on designing EDAs based on
  some new models, which are expected to be more powerful and less
  time-consuming to build.

  3. MRI Magnet Design  

  Magnetic Resonance Imaging is a technique widely used in medical
  environments to produce high quality images of the inside of the
  human body.

  One of the key components of a MRI system is the magnet,which is
  used to generate an intense and very homogeneous field within a
  spherical area.

  In this project, we will try to use various EAs and EDAs to solve
  two major challenging problems: the design of magnet for short MRI
  systems and the design of magnet for open MRI systems.

Biography:

  Mr.Bo Yuan received his B.E. in Software from Nanjing University of
  Science and Technology, P.R.China in 1998 and his Master degree with
  concentration on Evoluationary Computation from The University of
  Queensland in 2002.

  He is currently a PhD student with Complex and Intelligent Systems Group
  and a student member of IEEE and IEEE Neural Networks Society.

Type:

Ph.D confirmation

Contact:

Dr. Marcus Gallagher, seminar host (marcusg@itee.uq.edu.au))
or Guido Governatori (ITEE seminar co-ordinator) (guido@itee.uq.edu.au)

ITEE seminar web page: http://www.itee.uq.edu.au/~seminar

[All seminars]