A Computational Analysis of Properties and Limitations of Neural Networks: Toward New Parallel Architectures for Learning

Abstract

The goal of our work has been to develop a solid theoretical framework for the problem of learning from examples, in order to evaluate Neural Network architecture and develop new powerful parallel techniques and algorithms. Our approach was based on the formulation of the problem of learning from examples as a problem of approximation of multivariate functions from sparse data, in such a way as to take advantage of existing large body of results in function approximation theory and regularization. Our work has been successful beyond our original expectations at the time we wrote the proposal. We have developed a sizable body of theoretical results and applications. Several projects, many outside our own group, are now pursuing different aspects of the theory, and are developing algorithms and applying the technique to practical domains.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jan 01, 1992
Accession Number
ADA246156

Entities

People

  • R. Rivest
  • T. A. Poggio

Organizations

  • Massachusetts Institute of Technology

Tags

Communities of Interest

  • Autonomy
  • Energy and Power Technologies

DTIC Thesaurus Topics

  • Algorithms
  • Applied Mathematics
  • Artificial Intelligence
  • Computer Graphics
  • Computer Science
  • Computer Vision
  • Dimensionality Reduction
  • Information Processing
  • Neural Networks
  • Object Recognition
  • Probability Distributions
  • Random Variables
  • Recognition
  • Standards
  • Three Dimensional
  • Time Series Analysis
  • Two Dimensional

Fields of Study

  • Computer science

Readers

  • Neural Network Machine Learning.
  • Theoretical Analysis.

Technology Areas

  • AI & ML
  • AI & ML - Machine Learning Algorithms
  • AI & ML - Neural Networks