Academic Bibliography

 Search 200 years of publications by Ghent University researchers.
Advanced search
1 file | 3.28 MB
Add to list
  1. Search results
  2. Leveraging uncertainty estimation for...

Leveraging uncertainty estimation for trustworthy predictions in decision-making

Arthur Thuy (UGent) and Dries Benoit (UGent)
(2022) EURO 2022 : conference handbook and abstracts : 32nd European Conference on Operational Research (EURO XXXII). p.55-55
Author
Organization
Abstract
Recent successes across a variety of domains have led to the widespread deployment of neural networks in the field of operations research. Neural networks offer strong predictive performance but are notoriously difficult to interpret, leading to black-box models. As a result, they are poorly suited to be an essential component of larger decision support systems, which rely on trustworthy predictions. In this work, methods from the probabilistic deep learning literature are presented that address this issue by quantifying predictive uncertainty. Well-calibrated uncertainty estimates convey information about when a model’s output should (or should not) be trusted, and allow a system to reject decisions due to low confidence. We investigate the added value of the probabilistic methods applied to the task of knowledge tracing, a subfield of educational data mining. We find that they produce well-calibrated uncertainty estimates. Moreover, the methods effectively flag potentially incorrect predictions on shifted data, without compromising on predictive performance.

Downloads

  • Download
    (...).pdf
    • full text (Published version)
    • |
    • UGent only
    • |
    • PDF
    • |
    • 3.28 MB

Citation

Please use this url to cite or link to this publication:

MLA
Thuy, Arthur, and Dries Benoit. “Leveraging Uncertainty Estimation for Trustworthy Predictions in Decision-Making.” EURO 2022 : Conference Handbook and Abstracts : 32nd European Conference on Operational Research (EURO XXXII), 2022, pp. 55–55.
APA
Thuy, A., & Benoit, D. (2022). Leveraging uncertainty estimation for trustworthy predictions in decision-making. EURO 2022 : Conference Handbook and Abstracts : 32nd European Conference on Operational Research (EURO XXXII), 55–55.
Chicago author-date
Thuy, Arthur, and Dries Benoit. 2022. “Leveraging Uncertainty Estimation for Trustworthy Predictions in Decision-Making.” In EURO 2022 : Conference Handbook and Abstracts : 32nd European Conference on Operational Research (EURO XXXII), 55–55.
Chicago author-date (all authors)
Thuy, Arthur, and Dries Benoit. 2022. “Leveraging Uncertainty Estimation for Trustworthy Predictions in Decision-Making.” In EURO 2022 : Conference Handbook and Abstracts : 32nd European Conference on Operational Research (EURO XXXII), 55–55.
Vancouver
1.
Thuy A, Benoit D. Leveraging uncertainty estimation for trustworthy predictions in decision-making. In: EURO 2022 : conference handbook and abstracts : 32nd European Conference on Operational Research (EURO XXXII). 2022. p. 55–55.
IEEE
[1]
A. Thuy and D. Benoit, “Leveraging uncertainty estimation for trustworthy predictions in decision-making,” in EURO 2022 : conference handbook and abstracts : 32nd European Conference on Operational Research (EURO XXXII), Helsinki, Finland, 2022, pp. 55–55.
@inproceedings{01H5M95ZCPSK1WQS5DJYR7HDBZ,
  abstract     = {{Recent successes across a variety of domains have led to the widespread deployment of neural networks in the field of operations research. Neural networks offer strong predictive performance but are notoriously difficult to interpret, leading to black-box models. As a result, they are poorly suited to be an essential component of larger decision support systems, which rely on trustworthy predictions. In this work, methods from the probabilistic deep learning literature are presented that address this issue by quantifying predictive uncertainty. Well-calibrated uncertainty estimates convey information about when a model’s output should (or should not) be trusted, and allow a system to reject decisions due to low confidence. We investigate the added value of the probabilistic methods applied to the task of knowledge tracing, a subfield of educational data mining. We find that they produce well-calibrated uncertainty estimates. Moreover, the methods effectively flag potentially incorrect predictions on shifted data, without compromising on predictive performance.}},
  author       = {{Thuy, Arthur and Benoit, Dries}},
  booktitle    = {{EURO 2022 : conference handbook and abstracts : 32nd European Conference on Operational Research (EURO XXXII)}},
  isbn         = {{9789519525419}},
  language     = {{eng}},
  location     = {{Helsinki, Finland}},
  pages        = {{55--55}},
  title        = {{Leveraging uncertainty estimation for trustworthy predictions in decision-making}},
  url          = {{https://euro2022espoo.com/}},
  year         = {{2022}},
}