Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A note on incorrect inferences in non-binary qualitative probabilistic networks

Jack Storror Carter

TL;DR

The paper shows that the prevalent symmetry assumption for positive influence in non-binary Qualitative Probabilistic Networks is invalid, leading to incorrect inferences by standard edge-reversal and message-passing algorithms. It demonstrates, via counterexamples to the $S^{+}$/$S^{-}$ definitions and a space-shuttle-inspired example, that positive dependence defined through $FSD$ does not imply symmetric influences or reliable reverse-inference in non-binary settings. The authors discuss remedies such as binarization, redefining positive dependence (potentially via MLRP or symmetry-aware definitions), and the need for cautious interpretation of inferences in non-binary QPNs, with ties to broader positive-dependence literature (e.g., MTP2). These findings underscore the practical impact on QPN-based reasoning and suggest concrete directions for robust future work.

Abstract

Qualitative probabilistic networks (QPNs) combine the conditional independence assumptions of Bayesian networks with the qualitative properties of positive and negative dependence. They formalise various intuitive properties of positive dependence to allow inferences over a large network of variables. However, we will demonstrate in this paper that, due to an incorrect symmetry property, many inferences obtained in non-binary QPNs are not mathematically true. We will provide examples of such incorrect inferences and briefly discuss possible resolutions.

A note on incorrect inferences in non-binary qualitative probabilistic networks

TL;DR

The paper shows that the prevalent symmetry assumption for positive influence in non-binary Qualitative Probabilistic Networks is invalid, leading to incorrect inferences by standard edge-reversal and message-passing algorithms. It demonstrates, via counterexamples to the / definitions and a space-shuttle-inspired example, that positive dependence defined through does not imply symmetric influences or reliable reverse-inference in non-binary settings. The authors discuss remedies such as binarization, redefining positive dependence (potentially via MLRP or symmetry-aware definitions), and the need for cautious interpretation of inferences in non-binary QPNs, with ties to broader positive-dependence literature (e.g., MTP2). These findings underscore the practical impact on QPN-based reasoning and suggest concrete directions for robust future work.

Abstract

Qualitative probabilistic networks (QPNs) combine the conditional independence assumptions of Bayesian networks with the qualitative properties of positive and negative dependence. They formalise various intuitive properties of positive dependence to allow inferences over a large network of variables. However, we will demonstrate in this paper that, due to an incorrect symmetry property, many inferences obtained in non-binary QPNs are not mathematically true. We will provide examples of such incorrect inferences and briefly discuss possible resolutions.
Paper Structure (6 sections, 1 theorem, 1 equation, 3 figures)

This paper contains 6 sections, 1 theorem, 1 equation, 3 figures.

Table of Contents

  1. Introduction
  2. QPNs
  3. Positive dependence
  4. Inference in QPNs
  5. Example
  6. Discussion and future work

Key Result

Proposition 1

$X$ and $Y$ satisfy the MLRP if and only if $X$ positively influences $Y$ for any choice of prior distribution on $Y$.

Figures (3)

  • Figure 1: An example QPN.
  • Figure 2: A very simple QPN in which $X$ positively influences $Y$. However, as demonstrated in Section \ref{['sec:PD']}, we cannot conclude that $Y$ also positively influences $X$.
  • Figure 3: Left: A QPN for space shuttle example. Right: Results of a message passing algorithm after observing a large value for HeOx Temp Probe. Both figures originally appeared in Druzdzel1993.

Theorems & Definitions (4)

  • Definition 1
  • Definition 2
  • Definition 3
  • Proposition 1