143. Parameterization Support of Bayesian Networks Using a Large Language Model
Invited abstract in session WE-7: Simulation, Data & Decision Support, stream Simulation and Quantum Computing.
Wednesday, 16:30-18:00Room: U2-205
Authors (first author is the speaker)
| 1. | Max Krueger
|
| Computer Science, Technische Hochschule Ingolstadt |
Abstract
In recent decades, Bayesian Networks have been used in a variety of applications, particularly for classification problems. When building Bayesian Networks, in addition to creating a suitable qualitative structure, i.e., a Directed Acyclic Graph (DAG), a quantitative parameterization of the conditional probabilities must also be conducted for all nodes. Each node of the DAG represents an aspect of the application as a random variable. If no training data is available for parameter estimation when using learning methods for Bayesian Networks, subject matter experts are usually consulted to estimate the required conditional probabilities based on their application knowledge. The idea behind this contribution is to replace the support of these application experts, who may be challenging to find. The parameterization of Bayesian Networks is then condcuted by application of a large language model. We describe a possible approach in which the necessary parameters of the Bayesian network classifier are to be determined for an example application from air surveillance with the help of the LLM-based chatbot ChatGPT. Finally, the resulting parameterization of the Bayesian Network generated in this way is compared with a classification network generated by subject matter experts with regard to its classification performance.
Keywords
- Decision Support Systems
- Military Operations Research
- Simulation
Status: accepted
Back to the list of papers