Causal approximations

  • Authors:

  • P. Pandurang Nayak

  • Affiliations:

  • Knowledge Systems Laboratory, Stanford University, Palo Alto, CA

  • Venue:
  • AAAI'92 Proceedings of the tenth national conference on Artificial intelligence
  • Year:
  • 1992

Quantified Score

Hi-index 0.01

Visualization

Abstract

Adequate problem representations require the identification of abstractions and approximations that are well suited to the task at hand. In this paper we introduce a new class of approximations, called causal approximations, that are commonly found in modeling the physical world. Causal approximations support the efficient generation of parsimonious causal explanations, which play an important role in reasoning about engineered devices. The central problem to be solved in generating parsimonious causal explanations is the identification of a simplest model that explains the phenomenon of interest. We formalize this problem and show that it is, in general, intractable. In this formalization, simplicity of models is based on the intuition that using more approximate models of fewer phenomena leads to simpler models. We then show that when all the approximations are causal approximations, the above problem can be solved in polynomial time.