Given a collection of objects in the world, along with some relations that hold among them, a fundamental problem is how to learn definitions of some relations and concepts of interest in terms of the given relations. These definitions might be quite complex and, inevitably, might require the use of quantified expressions. Attempts to use first order languages for these purposes are hampered by the fact that relational inference is intractable and, consequently, so is the problem of learning relational definitions. This work develops an expressive relational representation language that allows the use of propositional learning algorithms when learning relational definitions.

The representation serves as an intermediate level between a raw description of observations in the world and a propositional learning system that attempts to learn definitions for concepts and relations. It allows for hierarchical composition of relational expressions that can be evaluated efficiently on the observations and thus supports learning complex definitions by learning simple functions of the intermediate representations. The approach is illustrated using examples from natural language and visual processing.

@conference{CumbyRo00,
  author = {C. Cumby and D. Roth},
  title = {Relational Representations that facilitate learning},
  booktitle = {KR},
  pages = {425--434},
  year = {2000},
  acceptance = {62/172 (36\%)},
  url = " http://cogcomp.cs.illinois.edu/papers/kr00.pdf",
  funding = {NSF98,KDI},
  projects = {LT,KR},
  comment = {FEX: feature extraction language; relational features; relational generation functions},
}