Take Care When Tuning Bayesian Networks

Janneke H. Bolt; Arjen Hommersom; Silja Renooij

doi:10.1007/978-3-032-29000-7_33

Take Care When Tuning Bayesian Networks

Janneke H. Bolt^*
, Arjen Hommersom
, Silja Renooij

^*Corresponding author for this work

Sub Intelligent Systems

Open University of the Netherlands

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

Abstract

Parameter tuning in Bayesian networks aims to fine-tune certain output probabilities to ensure that they match the expected outcome based upon domain expertise. Various tuning heuristics exist that can change different sets of parameters to arrive at the requested change in output. Previous research has evaluated the performance of these heuristics by comparing networks before and after tuning. In this paper, we study the performance of tuning heuristics by qualitatively inspecting the changes they incur compared to a ground truth distribution. We show for example networks with a fixed structure that tuning an output probability towards its ground truth value does not necessarily mean that a network as a whole more closely fits the ground truth. In fact, this fit can become worse; tuning of Bayesian networks should therefore be done with care.

Original language	English
Title of host publication	Information Processing and Management of Uncertainty in Knowledge-Based Systems - 21st International Conference, IPMU 2026, Proceedings
Editors	Barbara Vantaggi, Davide Petturiti, Giulianella Coletti, Thierry Denoeux, Anne Laurent, Enrique Miranda, Jesús Medina, Bernadette Bouchon-Meunier, Ronald R. Yager
Publisher	Springer
Pages	469-481
Number of pages	13
ISBN (Electronic)	978-3-032-29000-7
ISBN (Print)	978-3-032-28999-5
DOIs	https://doi.org/10.1007/978-3-032-29000-7_33
Publication status	Published - 12 Jun 2026
Event	21st International Conference on Information Processing and Management of Uncertainty in Knowledge-Based Systems, IPMU 2026 - Rome, Italy Duration: 15 Jun 2026 → 19 Jun 2026

Publication series

Name	Communications in Computer and Information Science
Volume	3021 CCIS
ISSN (Print)	1865-0929
ISSN (Electronic)	1865-0937

Conference

Conference	21st International Conference on Information Processing and Management of Uncertainty in Knowledge-Based Systems, IPMU 2026
Country/Territory	Italy
City	Rome
Period	15/06/26 → 19/06/26

Bibliographical note

Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.

Keywords

Bayesian networks
Ground truth comparison
Parameter tuning

Access to Document

10.1007/978-3-032-29000-7_33

Cite this

Bolt, J. H., Hommersom, A., & Renooij, S. (2026). Take Care When Tuning Bayesian Networks. In B. Vantaggi, D. Petturiti, G. Coletti, T. Denoeux, A. Laurent, E. Miranda, J. Medina, B. Bouchon-Meunier, & R. R. Yager (Eds.), Information Processing and Management of Uncertainty in Knowledge-Based Systems - 21st International Conference, IPMU 2026, Proceedings (pp. 469-481). (Communications in Computer and Information Science; Vol. 3021 CCIS). Springer. https://doi.org/10.1007/978-3-032-29000-7_33

Bolt, Janneke H. ; Hommersom, Arjen ; Renooij, Silja. / Take Care When Tuning Bayesian Networks. Information Processing and Management of Uncertainty in Knowledge-Based Systems - 21st International Conference, IPMU 2026, Proceedings. editor / Barbara Vantaggi ; Davide Petturiti ; Giulianella Coletti ; Thierry Denoeux ; Anne Laurent ; Enrique Miranda ; Jesús Medina ; Bernadette Bouchon-Meunier ; Ronald R. Yager. Springer, 2026. pp. 469-481 (Communications in Computer and Information Science).

@inproceedings{d40aa079016e4a689ae109d65d02e15e,

title = "Take Care When Tuning Bayesian Networks",

abstract = "Parameter tuning in Bayesian networks aims to fine-tune certain output probabilities to ensure that they match the expected outcome based upon domain expertise. Various tuning heuristics exist that can change different sets of parameters to arrive at the requested change in output. Previous research has evaluated the performance of these heuristics by comparing networks before and after tuning. In this paper, we study the performance of tuning heuristics by qualitatively inspecting the changes they incur compared to a ground truth distribution. We show for example networks with a fixed structure that tuning an output probability towards its ground truth value does not necessarily mean that a network as a whole more closely fits the ground truth. In fact, this fit can become worse; tuning of Bayesian networks should therefore be done with care.",

keywords = "Bayesian networks, Ground truth comparison, Parameter tuning",

author = "Bolt, \{Janneke H.\} and Arjen Hommersom and Silja Renooij",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.; 21st International Conference on Information Processing and Management of Uncertainty in Knowledge-Based Systems, IPMU 2026 ; Conference date: 15-06-2026 Through 19-06-2026",

year = "2026",

month = jun,

day = "12",

doi = "10.1007/978-3-032-29000-7\_33",

language = "English",

isbn = "978-3-032-28999-5",

series = "Communications in Computer and Information Science",

publisher = "Springer",

pages = "469--481",

editor = "Barbara Vantaggi and Davide Petturiti and Giulianella Coletti and Thierry Denoeux and Anne Laurent and Enrique Miranda and Jes{\'u}s Medina and Bernadette Bouchon-Meunier and Yager, \{Ronald R.\}",

booktitle = "Information Processing and Management of Uncertainty in Knowledge-Based Systems - 21st International Conference, IPMU 2026, Proceedings",

}

Bolt, JH, Hommersom, A & Renooij, S 2026, Take Care When Tuning Bayesian Networks. in B Vantaggi, D Petturiti, G Coletti, T Denoeux, A Laurent, E Miranda, J Medina, B Bouchon-Meunier & RR Yager (eds), Information Processing and Management of Uncertainty in Knowledge-Based Systems - 21st International Conference, IPMU 2026, Proceedings. Communications in Computer and Information Science, vol. 3021 CCIS, Springer, pp. 469-481, 21st International Conference on Information Processing and Management of Uncertainty in Knowledge-Based Systems, IPMU 2026, Rome, Italy, 15/06/26. https://doi.org/10.1007/978-3-032-29000-7_33

Take Care When Tuning Bayesian Networks. / Bolt, Janneke H.; Hommersom, Arjen; Renooij, Silja.
Information Processing and Management of Uncertainty in Knowledge-Based Systems - 21st International Conference, IPMU 2026, Proceedings. ed. / Barbara Vantaggi; Davide Petturiti; Giulianella Coletti; Thierry Denoeux; Anne Laurent; Enrique Miranda; Jesús Medina; Bernadette Bouchon-Meunier; Ronald R. Yager. Springer, 2026. p. 469-481 (Communications in Computer and Information Science; Vol. 3021 CCIS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - Take Care When Tuning Bayesian Networks

AU - Bolt, Janneke H.

AU - Hommersom, Arjen

AU - Renooij, Silja

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.

PY - 2026/6/12

Y1 - 2026/6/12

N2 - Parameter tuning in Bayesian networks aims to fine-tune certain output probabilities to ensure that they match the expected outcome based upon domain expertise. Various tuning heuristics exist that can change different sets of parameters to arrive at the requested change in output. Previous research has evaluated the performance of these heuristics by comparing networks before and after tuning. In this paper, we study the performance of tuning heuristics by qualitatively inspecting the changes they incur compared to a ground truth distribution. We show for example networks with a fixed structure that tuning an output probability towards its ground truth value does not necessarily mean that a network as a whole more closely fits the ground truth. In fact, this fit can become worse; tuning of Bayesian networks should therefore be done with care.

AB - Parameter tuning in Bayesian networks aims to fine-tune certain output probabilities to ensure that they match the expected outcome based upon domain expertise. Various tuning heuristics exist that can change different sets of parameters to arrive at the requested change in output. Previous research has evaluated the performance of these heuristics by comparing networks before and after tuning. In this paper, we study the performance of tuning heuristics by qualitatively inspecting the changes they incur compared to a ground truth distribution. We show for example networks with a fixed structure that tuning an output probability towards its ground truth value does not necessarily mean that a network as a whole more closely fits the ground truth. In fact, this fit can become worse; tuning of Bayesian networks should therefore be done with care.

KW - Bayesian networks

KW - Ground truth comparison

KW - Parameter tuning

UR - https://www.scopus.com/pages/publications/105042547279

U2 - 10.1007/978-3-032-29000-7_33

DO - 10.1007/978-3-032-29000-7_33

M3 - Conference contribution

AN - SCOPUS:105042547279

SN - 978-3-032-28999-5

T3 - Communications in Computer and Information Science

SP - 469

EP - 481

BT - Information Processing and Management of Uncertainty in Knowledge-Based Systems - 21st International Conference, IPMU 2026, Proceedings

A2 - Vantaggi, Barbara

A2 - Petturiti, Davide

A2 - Coletti, Giulianella

A2 - Denoeux, Thierry

A2 - Laurent, Anne

A2 - Miranda, Enrique

A2 - Medina, Jesús

A2 - Bouchon-Meunier, Bernadette

A2 - Yager, Ronald R.

PB - Springer

T2 - 21st International Conference on Information Processing and Management of Uncertainty in Knowledge-Based Systems, IPMU 2026

Y2 - 15 June 2026 through 19 June 2026

ER -

Bolt JH, Hommersom A, Renooij S. Take Care When Tuning Bayesian Networks. In Vantaggi B, Petturiti D, Coletti G, Denoeux T, Laurent A, Miranda E, Medina J, Bouchon-Meunier B, Yager RR, editors, Information Processing and Management of Uncertainty in Knowledge-Based Systems - 21st International Conference, IPMU 2026, Proceedings. Springer. 2026. p. 469-481. (Communications in Computer and Information Science). doi: 10.1007/978-3-032-29000-7_33

Take Care When Tuning Bayesian Networks

Abstract

Publication series

Conference

Bibliographical note

Keywords

Access to Document

Other files and links

Embargoed Document

Fingerprint

Cite this