Speaker
Description
Genetic algorithms and genetic progamming are metaheuristics taking inspi-
ration from Darwin’s theory of evolution to solve optimization problems. Ge-
netic algorithms explore a solution space to find solutions to problems while
genetic programming works in a program space to identify a program which
when executed will find an optimal solution to the problem. Both these
approaches have high runtimes when applied to complex problems and are
usually implemented using distributed computing in these instances. More
recently, genetic algorithms and genetic programming have been employed
by hyper-heuristics and have been used for the automated design of machine
learning and search techniques. Hyper-heuristics explore the heuristic space
rather than the solution space and hence search in the heuristic space is
mapped to the solution space. Automated design of machine learning and
search techniques is an emerging field aimed at removing the load of de-
sign, which is a time consuming task, from the researcher. This will also
enable non-experts to use machine learning toolkits that automate the de-
sign and hence allow the researcher to focus on the problem being solved.
The use of genetic algorithms and genetic programming in hyper-heuristics
and for automated design require additional processing time. The talk will
firstly look at the high performance computing architectures implemented
by our research group to reduce the runtimes of genetic programming and
genetic algorithms, particularly for hyper-heuristics and automated design.
An overview of some of the real-world problems that this has enabled us
to solve will then be presented. These include inducing human competitive
heuristics for solving timetabling problems, network intrusion detection in
the area of computer security, the automated design of techniques for finan-
cial forecasting, computer security, packing and logistics problems and the
introduction of multi-space search algorithms that perform search over more
than once space with applications in packing and forecasting.
