Learning Structure From Statistical Models

Statistical relational learning is a newly emerging area of machine learning that combines statisticalmodeling with relational representations. Here we argue that it provides a unified framework for the
discovery of structural information that can be exploited by a data management system. The categories
of structure that can be discovered include: instance-level dependencies and correlations, for example
intra-table column dependencies and inter-table join dependencies; record linkages and duplicates; and
schema matching and schema discovery from unstructured and semi-structured data.