Generating Conditional Probabilities for Bayesian Networks: Easing the Knowledge Acquisition Problem

TL;DR

This paper introduces an algorithm that simplifies the process of populating conditional probability tables in Bayesian networks by using weighted sums of elicited distributions, reducing cognitive load and leveraging information geometry.

Contribution

The paper presents a novel algorithm that eases knowledge acquisition for Bayesian networks by combining expert-elicited distributions with weighted sums, guided by information geometry.

Findings

Reduces the number of probability distributions needed from exponential to linear.

Uses weighted sums to accurately reflect expert judgments.

Employs information geometry to validate the probabilistic aggregation method.

Abstract

The number of probability distributions required to populate a conditional probability table (CPT) in a Bayesian network, grows exponentially with the number of parent-nodes associated with that table. If the table is to be populated through knowledge elicited from a domain expert then the sheer magnitude of the task forms a considerable cognitive barrier. In this paper we devise an algorithm to populate the CPT while easing the extent of knowledge acquisition. The input to the algorithm consists of a set of weights that quantify the relative strengths of the influences of the parent-nodes on the child-node, and a set of probability distributions the number of which grows only linearly with the number of associated parent-nodes. These are elicited from the domain expert. The set of probabilities are obtained by taking into consideration the heuristics that experts use while arriving at probabilistic estimations. The algorithm is used to populate the CPT by computing appropriate weighted sums of the elicited distributions. We invoke the methods of information geometry to demonstrate how these weighted sums capture the expert's judgemental strategy.