University of Illinois Urbana-Champaign

Functional Subsumption in Feature Description Logic

Braz, Rodrigo de Salvo; Roth, Dan

Content Files
Functional Subsumption in Feature Description Logic.pdf

Permalink

https://hdl.handle.net/2142/10802

Description

Title
Functional Subsumption in Feature Description Logic
Author(s)
  • Braz, Rodrigo de Salvo
  • Roth, Dan
Issue Date
2004-04
Keyword(s)
Machine Learning
Date of Ingest
2009-04-14T20:44:30Z
Abstract
Most machine learning algorithms rely on examples represented propositionally as feature vectors. However, most data in real applications is structured and better described by sets of objects with attributes and relations between them. Typically, ad-hoc methods have been used to convert such data to feature vectors, taking, in many cases, a significant amount of the computation. Propositionalization becomes more principled if generating features from structured data is done using a formal, domain independent language that describes feature types to be extracted. This language must have limited expressivity in order to be useful while some inference procedures on it are still tractable. In this chapter we present Feature Description Logic (FDL), proposed by Cumby & Roth, where feature extraction is viewed as an inference process (subsumption). We also present an extension to FDL we call Functional FDL. FDL is ultimately based on the unification of object attributes and relations between objects in order to detect feature types in examples. Functional subsumption provides further abstraction by using unification modulo a Boolean function representing similarity between attributes and relations. This greatly improves flexibility in practical situations by accomodating variations in attributes values and relation names, incorporating background knowledge (e.g., typos, number and gender, synomyms etc). We define the semantics of Functional subsumption and how to adapt the regular subsumption algorithm for implementing it.
Type of Resource
text
Genre of Resource
Technical Report
Permalink
http://hdl.handle.net/2142/10802
Copyright and License Information
You are granted permission for the non-commercial reproduction, distribution, display, and performance of this technical report in any format, BUT this permission is only for a period of 45 (forty-five) days from the most recent time that you verified that this technical report is still available from the University of Illinois at Urbana-Champaign Computer Science Department under terms that include this permission. All other rights are reserved by the author(s).

Owning Collections

Research and Tech Reports - Computer Science PRIMARY