Never-Ending Learning
Abstract
The goal of this research was to explore a new approach to machine learning, called Never-Ending Learning. Although machine learning research has been increasingly successful in recent years, this effort addressed developing a machine learning system that learns cumulatively forever, using what was learned yesterday to improve its ability to learn tomorrow, and improving daily, indefinitely. The thesis underlying this research is that the vast redundancy of information on the web (e.g., many facts are stated multiple times in different ways) will enable a system with the right learning mechanisms and capabilities for self-reflection to learn with only occasional outside supervision. A general approach is described for building a never-ending language learner that uses semi-supervised learning methods, an ensemble of varied knowledge extraction methods, and a flexible knowledge base representation that allows the integration of the outputs of those methods.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 2010
- Accession Number
- ADA527602
Entities
People
- Tom M. Mitchell
Organizations
- Carnegie Mellon University