TITLE: Towards using Grammatical Evolution to evolve complex behaviours Towards using Grammatical Evolution to evolve complex behaviours
PRESENTER: Robin Harper, http://www.cse.unsw.edu.au/db/staff/info/robinh.html, email@example.com.EDU.AU
AFFILIATION:CSE - UNSW, http://www.cse.unsw.edu.au
DATE: Monday 12th September 2005
PLACE: CSE Seminar Room Level 1 K17
Grammatical Evolution (GE) is a method of utilising an evolutionary
algorithm to evolve code written in an arbitrary language, provided
the grammar for that language can be expressed in a Backus Naur Form.
GE pulls together various concepts relating to genetic programming,
namely the idea of evolving programs [Koza], the benefits of typing
[Montana][Whigham] (which arise as a consequence of constraining the
syntax of the evolved program to a that specified in the BNF grammar
utilised) and the advantages of separating the underlying
representation (the genotype) from the evolved program (the phenotype)
This thesis aims to investigate:
1. the benefits of allowing the grammar to dynamically define and use
functions (in an attempt to allow modularity to be evolved); and
2. the utility of alternative methods of searching the geno-space,
with a view to allowing the selection of the search strategy to be brought
under control of the evolutionary process; and
3. the implementation of a productive method of allowing the mutation
rate to be dynamically modified; and
4. the effect radically different phenotype representations
(program/L-Systems/edge encoded neural nets) have on the ability of GE
to find different solutions.
Assuming a sufficiently robust evolutionary system can be produced, it
will be used to evolve agents in a complex artificial world in an
attempt to evolve complex emergent behaviours.
BIOGRAPHY OF SPEAKER:
Robin Harper is doing Master by Research at CSE under supervision of Dr Alan Blair. His thesis topic is about Evolutionary Computation/Adaptive Agents/AI.
Van Hai Ho