Abstract
M.Sc.
Keywords: Multi-agent systems, cellular automata, cellular evolutionary
algorithms, overlay networks, knowledge representation, prior art detection.
The dissertation explores a set of naturally-inspired meta-heuristics and their
application to the improvement of multi-agent based prior art detection
techniques within the corpus of patent repositories.
The discussion begins with an examination of the family of emergent search
algorithms which draw their inspiration from Darwinian evolution. These
evolutionary algorithms are examined in terms of the increasing
expressiveness of their representations from simple strings, through trees to
separate genomic and phenotypic representations.
The discussion then examines a class of computational entities known as
cellular automata which draw their inspiration from crystalline growth patterns.
Cellular automata are examined in terms of their underlying topologies and
emergent properties in particular those of computational irreducibility and
universality. Cellular automata are related to evolutionary algorithms as both
an end product of evolution and as a driver for it.
Software agency is then presented as a logical successor to the objectoriented
software paradigm. The notion of rational agency is explored in both
the single and multi-agent case and then coupled with evolutionary algorithm
research.
The field of information retrieval is investigated. Particular attention is paid to
the imposition of structure onto inherently unstructured textual information.
A variety of overlay networks are explored as a basis for the distributed
storage of information. The JADE agent development framework is analysed
as a special case of overlay network as well as a container for intelligence.
The previously introduced topics are then combined into a novel technique for
prior art detection in distributed patent repositories. The task of patent mining
is decomposed into an agent-oriented model based on subsumption
hierarchies. A resource description framework based patent application
representation is developed together with the use of appropriate algorithms to
analyse it in terms of both structural and content based similarity against a
given corpus of patents.
A novel multi-agent oriented cellular gene expression system is developed
and implemented to improve the effectiveness of the analysis algorithms. Of
particular interest is the use of universal elementary cellular automata to
coordinate the interaction of individual evolutionary agents. Every possible
elementary cellular automaton is evaluated for suitability and the structural
analysis algorithm's efficiency is explored for both human and machine
generated patents.