Evaluating model calibration in classification
By Juozas Vaicenavicius, David Widmann, Carl R. Andersson, Fredrik Lindsten, Jacob Roll, Thomas Bo Schön. 2019
273 citations recorded by the source corpus.
Probabilistic classifiers output a probability distribution on target classes rather than just a class prediction. Besides providing a clear separation of prediction and decision making, the main advantage of probabilistic models is their ability to represent uncertainty about predictions. In safety-critical applications, it is pivotal for a model to possess an adequate sense of uncertainty, which for probabilistic classifiers translates into outputting probability distributions that are consistent with the empirical frequencies observed from realized outcomes. A classifier with such a property is called calibrated. In this work, we develop a general theoretical calibration evaluation framework grounded in probability theory, and point out subtleties present in model calibration evaluation that lead to refined interpretations of existing evaluation techniques. Lastly, we propose new ways to quantify and visualize miscalibration in probabilistic classification, including novel multidimensional reliability diagrams.
Subfields: Software
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Software
Citations in the corpus
- Verified Uncertainty Calibration (2019, neutral): “This result is similar to Theorem 2 in recent work [28].”
- Measuring Calibration in Deep Learning (2019, background): “Specifically, calibration is a guarantee for all predicted probabilities p ∈ [0, 1] and class labels y ∈ {0, . . . ,K − 1} (see also the calibration function of Vaicenavicius et al. (2019)).”
- Non-Parametric Calibration for Classification (2019, background): “A theoretical framework for evaluating calibration in classification was suggested by Vaicenavicius et al. (2019).”
- Calibration tests in multi-class classification: A unifying framework (2019, methodology): “Recently Vaicenavicius et al. (2019) suggested that this analysis might be too reduced for many realistic scenarios.”
- Beyond temperature scaling: Obtaining well-calibrated multiclass probabilities with Dirichlet calibration (2019, methodology): “One can define several weaker notions of calibration [25] which provide necessary conditions for the model to be fully calibrated.”
- A survey of uncertainty in deep neural networks (2021, methodology): “consistency between the predictive distributions and the observations [286].”
- Calibrating Deep Neural Networks using Focal Loss (2020, background): “A stronger definition of calibration would require the probabilities of all the classes in the softmax distribution to be calibrated [14, 36, 39, 15].”
- Evaluating Scalable Bayesian Deep Learning Methods for Robust Computer Vision (2019, methodology): “As an absolute measure of uncertainty estimation quality, we therefore also evaluate the methods in terms of calibration [5, 46].”
- Pitfalls of In-Domain Uncertainty Estimation and Ensembling in Deep Learning (2020, background): “Assuming B m denotes the m -th bin and M is overall number of bins, the ECE is defined as follows: where A recent line of works on measuring calibration in deep learning (Vaicenavicius et al., 2019; Kumar et al., 2019; Nixon et al., 2019) outlines several problems of the ECE score.”
- Revisiting the Calibration of Modern Neural Networks (2021, background): “Nevertheless, it is well understood that estimating ECE accurately is difficult because estimators can be strongly biased and many estimator variants exist (Nixon et al., 2019; Roelofs et al., 2020; Vaicenavicius et al., 2019; Gupta et al., 2021).”
- Classification with Valid and Adaptive Coverage (2020, background): “A different but related line of work focuses on post-processing the output of black-box classification algorithms to produce more accurate probability estimates [6, 11, 15, 16, 22, 27, 28], although without achieving prediction sets with provable finite-sample coverage.”