Testing for Learning with Small Data Sets.

Abstract

The objective of this research was to develop a simple methodology to test for learning using a small sample size and to develop a procedure for measuring the rate of learning at any particular trial. For this research the time between trials was considered insignificant in affecting previously gained knowledge. Two linear methods and one nonlinear method were developed to test for learning by examining the rate of learning over several trials. In the nonlinear method, estimates for sigma sub epsilon and the parameters 'a' and 'b' are obtained and a test on the degree of nonlinearity of the function is conducted using Beales measure of nonlinearity. If the degree of nonlinearity is small enough then the confidence interval for the slope at any trial can be evaluated by using linear theory approximations. In a comparison of the two procedures, the linear methods were more powerful tests, however, the nonlinear method was able to provide information on the rate of learning at each trial when the nonlinearity conditions were satisfied and significant learning was detected. The more powerful linear test procedure was the LLSR method, which can detect an average rate of learning over 15 trials of .01 at an alpha = .05 level 95% of the time when the standard deviation is sigma sub epsilon < or = .05.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jun 01, 1978
Accession Number
ADA086176

Entities

People

  • Kenneth Alan Yealy

Organizations

  • Georgia Tech

Tags

Communities of Interest

  • C4I
  • Energy and Power Technologies
  • Ground and Sea Platforms
  • Weapons Technologies

DTIC Thesaurus Topics

  • Computational Science
  • Computer Programs
  • Computer Simulations
  • Computers
  • Confidence Limits
  • Data Science
  • Data Sets
  • Estimators
  • Information Science
  • Plastic Explosives
  • Probability
  • Simulations
  • Standards
  • Statistical Tests
  • Statistics
  • Test And Evaluation
  • Test Methods

Readers

  • Neural Network Machine Learning.
  • Operations Research
  • Regression Analysis.