A noise metric on binary training inputs and a framework for learning generalization

Ahmet Ugur, Henry Thompson

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Training is an essential phase for a supervised learning algorithm. Testing phase is equally important to assess the degree of learning. In this paper, we discuss a noise level measure on binary input patterns with respect to a training set. The input space can then be partitioned into groups of inputs representing different degrees of noise with respect to the training set used Test inputs can be selected from all partitions. A normalized response of the processor being trained to test inputs from all partitions is defined. This response represents the overall response of the processor to the set of all inputs, and is the generalization trend of the processor. The representation of the overall response opens up the possibility of learning a particular generalization after the training phase. A general algorithm is presented to learn a particular generalization and potential issues are discussed.

Original languageEnglish
Title of host publicationProceedings of the International Conference on Artificial Intelligence, IC-AI'04 and Proceedings of the International Conference on Machine Learning; Models, Technologies and Applications, MLMTA'04)
EditorsH.R. Arabnia, M. Youngsong
Pages911-915
Number of pages5
StatePublished - 2004
EventProceedings of the International Conference on Artificial Intelligence, IC-AI'04 - Las Vegas, NV, United States
Duration: Jun 21 2004Jun 24 2004

Publication series

NameProceedings of the International Conference on Artificial Intelligence, IC-AI'04
Volume2

Conference

ConferenceProceedings of the International Conference on Artificial Intelligence, IC-AI'04
Country/TerritoryUnited States
CityLas Vegas, NV
Period06/21/0406/24/04

Keywords

  • Learning generalization
  • Noise level partitioning
  • Supervised learning

Fingerprint

Dive into the research topics of 'A noise metric on binary training inputs and a framework for learning generalization'. Together they form a unique fingerprint.

Cite this