AASERT: Dynamic Training of Humans and Tutoring Agents

Abstract

Our ONR funded research over the past several years has focused on how machine learning systems can continuously improve through the dynamic modification of architecture and the dynamic construction of training environments. Although for different applications we use many different formalisms (neural networks, genetic programs, adaptive dynamical systems), we have focused on a framework for learning in which the environment automatically and incrementally becomes more challenging as the learner progresses. Inspired by "arms race" and sexual selection phenomena in natural evolution, we call this 'co-evolutionary learning'. It involves a set of learners competing in a game, learning from each other, without a teacher. This AASERT award will take one additional graduate student into an exploration of whether the principles discovered in our machine learning work could be the basis for a new kind of educational technology, where students provide appropriate challenges to each other across the internet, reducing the need for teacher supervision.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Mar 31, 2001
Accession Number
ADA405402

Entities

People

  • Jordan Pollack

Organizations

  • Brandeis University

Tags

DTIC Thesaurus Topics

  • Artificial Intelligence
  • Cognitive Science
  • Computer Programs
  • Computer Science
  • Computers
  • Education
  • Educational Technology
  • Internet
  • Machine Learning
  • Mathematical Analysis
  • Neural Networks
  • Psychological Phenomena And Processes
  • Psychology
  • Social Psychology
  • Students
  • Training
  • Video Games

Fields of Study

  • Computer science

Readers

  • Artificial Intelligence
  • Neural Network Machine Learning.
  • Research Science/Academic Research

Technology Areas

  • AI & ML
  • Biotechnology