Parameterized Temperature Scaling for Boosting the Expressive Power in Post-Hoc Uncertainty Calibration

Tomani C, Cremers D, Buettner F (2022)

Publication Type: Conference contribution

Publication year: 2022

Journal

Lecture Notes in Computer Science Springer Verlag

Publisher: Springer Science and Business Media Deutschland GmbH

Book Volume: 13673 LNCS

Pages Range: 555-569

Conference Proceedings Title: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Event location: Tel Aviv, ISR

ISBN: 9783031197772

DOI: 10.1007/978-3-031-19778-9_32

Abstract

We address the problem of uncertainty calibration and introduce a novel calibration method, Parametrized Temperature Scaling (PTS). Standard deep neural networks typically yield uncalibrated predictions, which can be transformed into calibrated confidence scores using post-hoc calibration methods. In this contribution, we demonstrate that the performance of accuracy-preserving state-of-the-art post-hoc calibrators is limited by their intrinsic expressive power. We generalize temperature scaling by computing prediction-specific temperatures, parameterized by a neural network. We show with extensive experiments that our novel accuracy-preserving approach consistently outperforms existing algorithms across a large number of model architectures, datasets and metrics (Source code available at: https://github.com/tochris/pts-uncertainty ).

Involved external institutions

Technische Universität München (TUM) DE Germany (DE) Deutsches Krebsforschungszentrum (DKFZ) DE Germany (DE)

How to cite

APA:

Tomani, C., Cremers, D., & Buettner, F. (2022). Parameterized Temperature Scaling for Boosting the Expressive Power in Post-Hoc Uncertainty Calibration. In Shai Avidan, Gabriel Brostow, Moustapha Cissé, Giovanni Maria Farinella, Tal Hassner (Eds.), Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (pp. 555-569). Tel Aviv, ISR: Springer Science and Business Media Deutschland GmbH.

MLA:

Tomani, Christian, Daniel Cremers, and Florian Buettner. "Parameterized Temperature Scaling for Boosting the Expressive Power in Post-Hoc Uncertainty Calibration." Proceedings of the 17th European Conference on Computer Vision, ECCV 2022, Tel Aviv, ISR Ed. Shai Avidan, Gabriel Brostow, Moustapha Cissé, Giovanni Maria Farinella, Tal Hassner, Springer Science and Business Media Deutschland GmbH, 2022. 555-569.

BibTeX: Download