Beyond Sigmoids : : How to obtain well-calibrated probabilities from binary classifiers with beta calibration
- Submitting institution
-
University of Bristol
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 148218393
- Type
- D - Journal article
- DOI
-
10.1214/17-EJS1338SI
- Title of journal
- Electronic Journal of Statistics
- Article number
- -
- First page
- 5052
- Volume
- 11
- Issue
- 2
- ISSN
- 1935-7524
- Open access status
- Compliant
- Month of publication
- December
- Year of publication
- 2017
- URL
-
-
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
2
- Research group(s)
-
A - Artificial Intelligence and Autonomy
- Citation count
- 7
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- Calibrated probability estimates are crucial for optimal and trustworthy decision making. In this work we generalise the commonly used logistic calibration method to correct over-confident classifiers, e.g. deep neural networks. The work on which this paper is based was selected for plenary presentation at AISTATS, an important and highly selective machine learning conference, and this significantly extended paper was invited and fast-tracked for an EJS special issue. Part of an international collaboration funded by CHIST-ERA (https://www.chistera.eu/projects/reframe, 2012-15, €950k for Bristol-Valencia-Strasbourg), leading to NeurIPS'19 paper (26 citations in less than a year) extending the approach to more than two classes.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -