Jukka Hekanaho

Learn More
We describe a GA based concept learning theory revision system DOGMA and discuss how it can be applied to relational learning The search for better theories in DOGMA is guided by a novel tness function that combines the minimal description length and information gain measures To show the e cacy of the system we compare it to other learners in three(More)
We study the integration of background knowledge and concept learning genetic algorithms and show how they have been integrated in the system DOGMA Our emphasis is in speeding up the inductive learning process by using suggestions from the background knowledge to direct genetic search We don t do theory revision by patching the old theory rather we build a(More)
We describe an application of DOGMA, a GA-based theory revision system, to MDL-based rule enhancement in supervised concept learning. The system takes as input classi cation data and a rule-based classi cation theory, produced by some rule-based learner, and builds a second, hopefully more accurate, model of the data. Unlike most theory revision systems(More)
Acknowledgements As my thesis is now written, it is time to pause for a while, to take a look back and thank all the people who have been involved in the process. I would like to start by thanking my supervisor Professor Ralph Back. He has had long-sightedness in allowing me to do what I wanted to do and has had patience with the long incubation of my(More)
Sharing is a popular method for introducing divergency in genetic algorithms for multimodal function optimization In this paper we de ne three sharing methods and investigate their applicability for concept learning of disjuncti ve rules from preclassi ed examples The sharing methods are incorporated in JGA a genetic algorithm based concept learning system(More)
We describe how proof rules for three advanced reenement features are mechanically veriied using the HOL theorem prover. These features are data reenement, backwards data reenement and superposition reenement of initialised loops. We also show how applications of these proof rules to actual program reenement can be checked using the HOL system, with the HOL(More)
We study the use of genetic algorithms in rule based concept learning The developed system JGA is capable of learning disjunctive concepts in First Order Logic We take a two leveled approach that combines features from both the Michigan and Pittsburgh approaches We compare the system in several propositional domains with three well known concept learners(More)