Algorithms for identifying and learning under conditional distributions

Abstract

This proposal studies fundamental questions about search problems formulated by data distributions and incomplete information or background knowledge using a probabilistic semantics, known as PAC-semantics, for logical presentations. This data-driven formalism captures inductive generalization as opposed to deduction, to allow reasoning in a partial information model. For instance, a key algorithmic problem in PAC-Semantics is to certify a logical formula as "almost" valid on the basis of examples drawn from the background distribution. This approach to learning and reasoning with incomplete information on data is very different from other logical, statistical, or probabilistic reasoning models.

Document Details

Document Type
DoD Grant Award
Publication Date
Mar 23, 2016
Source ID
FA95501510209

Entities

People

  • Brendan Juba

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • Washington University in St. Louis

Tags

Fields of Study

  • Computer science

Readers

  • Calculus or Mathematical Analysis
  • Neural Network Machine Learning.