Currently I'm working on a paper for the 2009 IEEE International Conference on Intelligent Computing and Intelligent Systems (ICIS 2009) to be held in Shanghai, China in November. Hopefully it will be accepted. By then, I might just have enough results for another paper, to be given at the 2010 Genetic and Evolutionary Computation Conference in Portland Oregon in July next year.
Here's the Latest and Greatest version of the abstract of the paper I'm working on. It will probably change a bit before the submission deadline, which is in, er, about 25 hours. gulp
Genetic Algorithms (GAs) based on evolutionary processes are powerful tools for finding optima in complex problems involving many considerations. These optima typically correspond to peaks in n-dimensional manifolds. Most of the work on optimising GAs in the past has been to tailor them by embedding domain knowledge for specific applications into the structure of the genome, rather than to examine the general case where no domain knowledge is assumed. We discuss the different parameters that describe GAs in the general case, go on to define a number of problem classes where different values for some of these parameters are already known to be most appropriate, and propose a method for determining the optimal parameters applicable to each problem class. The method is to use a classical GA to optimise non-classical GAs. We map the candidate GA parameters into a genome, to form a population of GAs as phenotypes. A manifold is formed by defining a fitness function based on the computational resources required by these non-classical GAs to solve a typical problem of the specified class. Optima are then determined using the usual GA methods of breeding, natural selection and mutation. Difficulties that may arise in implementation are discussed, along with proposed solutions.So I better stop blogging, and get back to writing the paper, hadn't I?
4 comments:
Occam's Law, applied to GA's computational resources? Interesting take on the problem. Least-energy solutions have been around for some time (I was debating them, for a commercial project, in 1999!), but I must admit, I've never seen them applied in this manner.
(Can I ask if you intend to examine 2-d manifolds, or 3-d ones?)
Good luck! :-)
Carolyn Ann
n-dimensional, where n >>1. Say 17 or so.
I'm a bit intrigued about the concept of "evolutionary processes" in software design. In the abstract, you mention phenotypes, mutation and natural selection, so it sounds like some kind of Darwinian software optimisation process. I'm an engineer, although not a software engineer, and I write quite a bit of software. In general terms, how do these evolutionary processes work? How do you introduce mutations? I probably won't understand the answer.
Oh, sorry - I totally forgot about those! (It's been far, far too long since I did any serious reading of mathematics.)
Neat! :-)
Carolyn Ann
Post a Comment