[Paper Review] Integrating Probabilistic, Taxonomic and Causal Knowledge in Abductive Diagnosis

Dekang Lin, Randy Goebel|arXiv (Cornell University)|Mar 27, 2013

Biomedical Text Mining and Ontologies5 citations

TL;DR

This paper proposes a novel abductive diagnosis framework that integrates probabilistic, causal, and taxonomic knowledge to generate explanations as chains of causation events rather than isolated hypotheses. It computes scenario probabilities from conditional probabilities of causation events and achieves polynomial complexity in knowledge base size, with exponential scaling only in the number of observations, offering a scalable alternative to traditional abduction methods.

ABSTRACT

We propose an abductive diagnosis theory that integrates probabilistic, causal and taxonomic knowledge. Probabilistic knowledge allows us to select the most likely explanation; causal knowledge allows us to make reasonable independence assumptions; taxonomic knowledge allows causation to be modeled at different levels of detail, and allows observations be described in different levels of precision. Unlike most other approaches where a causal explanation is a hypothesis that one or more causative events occurred, we define an explanation of a set of observations to be an occurrence of a chain of causation events. These causation events constitute a scenario where all the observations are true. We show that the probabilities of the scenarios can be computed from the conditional probabilities of the causation events. Abductive reasoning is inherently complex even if only modest expressive power is allowed. However, our abduction algorithm is exponential only in the number of observations to be explained, and is polynomial in the size of the knowledge base. This contrasts with many other abduction procedures that are exponential in the size of the knowledge base.

Motivation & Objective

• To develop a unified abductive diagnosis framework that combines probabilistic, causal, and taxonomic knowledge for improved diagnostic reasoning.
• To model causation not as isolated events but as chains of causation events forming coherent scenarios.
• To enable explanations at varying levels of detail and precision using taxonomic hierarchies.
• To reduce computational complexity by ensuring the algorithm is exponential only in the number of observations, not the knowledge base size.
• To provide a principled method for computing the probabilities of diagnostic scenarios from conditional probabilities of causation events.

Proposed method

• Represents a diagnosis as an occurrence of a chain of causation events that collectively explain all observed phenomena.
• Uses causal knowledge to justify conditional independence assumptions among causation events.
• Applies taxonomic knowledge to allow causation modeling at multiple levels of abstraction and observation description at varying granularities.
• Computes the probability of a diagnostic scenario as the product of conditional probabilities of its constituent causation events.
• Employs an abduction algorithm that is exponential only in the number of observations, making it efficient for large knowledge bases.
• Leverages structural constraints from causal and taxonomic hierarchies to prune the search space and improve scalability.

Experimental results

Research questions

• RQ1 How can probabilistic, causal, and taxonomic knowledge be systematically integrated into a single abductive diagnosis framework?
• RQ2 What is an effective representation of diagnostic explanations that captures causal chains rather than isolated causes?
• RQ3 How can varying levels of detail and precision in observations and causal models be supported within a unified framework?
• RQ4 Can the computational complexity of abductive reasoning be reduced by decoupling exponential growth from knowledge base size?
• RQ5 What is the correct method for computing the probability of a diagnostic scenario based on conditional probabilities of causation events?

Key findings

• The proposed framework models explanations as chains of causation events, enabling more coherent and contextually accurate diagnoses.
• The probability of a diagnostic scenario can be computed as the product of conditional probabilities of its causation events, ensuring probabilistic consistency.
• The abduction algorithm scales polynomially with the size of the knowledge base, making it efficient for large-scale applications.
• The algorithm's exponential complexity depends only on the number of observations, not the size of the knowledge base, significantly improving scalability.
• The integration of taxonomic knowledge allows for flexible modeling of causation and observation at multiple levels of abstraction.
• The approach outperforms traditional abduction methods in computational efficiency by avoiding exponential blowup in knowledge base size.