Statistical Inference and Causal Reasoning

Abstract

In this paper, we show how degrees of belief about causal predictions can be derived from statistics about the truth of properties over time. By using statistical information analogous to traditional non-statistical rules of causation and knowledge about a specific time point, predictions can be made about both the change and persistence of properties for the next time point with some degree of belief. We show how to incrementally compute this degree of belief by combining statistics conditioned on successively larger subsets of the reasoner's knowledge. Furthermore, we solve the qualification problem through a powerful heuristic that builds these subsets by considering properties with highest impact first. This heuristic ignores relatively unlikely, redundant, or unrelated properties when deriving a prediction, while directing the focus along causal chains. The iterative formula and this heuristic define an algorithm that produces predictions with quickly increasing confidence, allowing computational resources to trade off against-accuracy.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jan 01, 1989
Accession Number
ADA250543

Entities

People

  • Jay C. Weber

Organizations

  • University of Rochester

Tags

Communities of Interest

  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Accuracy
  • Algorithms
  • Artificial Intelligence
  • Causal Reasoning
  • Computations
  • Computer Science
  • Computers
  • Ignition
  • Ignition Systems
  • Probability
  • Qualifications
  • Reasoning
  • Statistical Inference
  • Statistics
  • Uncertainty
  • Universities

Fields of Study

  • Computer science

Readers

  • Artificial Intelligence
  • Regression Analysis.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Machine Learning Algorithms