Graph Based Concept Learning

Concept Learning is a Machine Learning technique in which the learning process is driven by providing positive and negative examples to the learner. From those examples, the learner builds a hypothesis (concept) that describes the positive examples and excludes the negative examples. Inductive Logic Programming (ILP) systems have successfully been used as concept learners. Examples of those are Foil (Quinlan and Cameron 1993) and Progol (Muggleton 1995). The main engine of these systems is based in first order logic. In this research we introduce a graph based relational concept learning system called SubdueCL, which through the experiments has shown that it is competitive with ILP systems in different types of domains. SubdueCL is an extension made to the Subdue (Cook and Holder 1994) system, which is an unsupervised graph based learner. The Subdue system takes as input a labeled graph and discovers substructures (sub-graphs) that compress the input graph, according to the minimum description length principle and represent structural concepts in the data. The main discovery algorithm is a computationally constrained beam search. The algorithm begins with the substructure matching a single vertex in the graph. Each iteration the algorithm selects the best substructure and incrementally expands the instances of the substructure. The algorithm searches for the best substructure until all possible substructures have been considered or the total amount of computation exceeds a given limit. Evaluation of each substructure is determined by how well the substructure compresses the description length of the input graph. The best substructure found by Subdue can be used to compress the input graph, which can then be input to another iteration of Subdue. After several iterations, Subdue builds a hierarchical description of the input data where later substructures are defined in terms of substructures discovered on previous iterations. Subdue has been applied to several domains including image analysis, CAD circuit analysis, chemical reaction chains, and artificiallygenerated databases (Cook and Holder 1994). The SubdueCL system is an extension to Subdue. It uses Subdue’s core functions to perform graph operations, but the learning process is different because it works as a