Measuring Calibration in Deep Learning

Abstract

Overconfidence and underconfidence in machine learning classifiers is measured by calibration: the degree to which the probabilities predicted for each class match the accuracy of the classifier on that prediction. How one measures calibration remains a challenge: expected calibration error, the most popular metric, has numerous flaws which we outline, and there is no clear empirical understanding of how its choices affect conclusions in practice, and what recommendations there are to counteract its flaws. In this paper, we perform a comprehensive empirical study of choices in calibration measures including measuring all probabilities rather than just the maximum prediction, thresholding probability values, class conditionality, number of bins, bins that are adaptive to the datapoint density, and the norm used to compare accuracies to confidences. To analyze the sensitivity of calibration measures, we study the impact of optimizing directly for each variant with recalibration techniques. Across MNIST, Fashion MNIST, CIFAR-10/100, and ImageNet, we find that conclusions on the rank ordering of recalibration methods is drastically impacted by the choice of calibration measure. We find that conditioning on the class leads to more effective calibration evaluations, and that using the L2 norm rather than the L1 norm improves both optimization for calibration metrics and the rank correlation measuring metric consistency. Adaptive binning schemes lead to more stablity of metric rank ordering when the number of bins vary, and is also recommended. We open source a library for the use of our calibration measures.

References in corpus (6)

• Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
• On Calibration of Modern Neural Networks
• Deep Anomaly Detection with Outlier Exposure
• Accurate Uncertainties for Deep Learning Using Calibrated Regression
• Training Confidence-calibrated Classifiers for Detecting Out-of-Distribution Samples
• Evaluating model calibration in classification

Cited by in corpus (34)

• Hands-on Bayesian Neural Networks -- a Tutorial for Deep Learning Users
• Simple and Principled Uncertainty Estimation with Deterministic Deep Learning via Distance Awareness
• Does deep learning model calibration improve performance in class-imbalanced medical image classification?
• Verified Uncertainty Calibration
• Revisiting the Calibration of Modern Neural Networks
• Label Leakage and Protection in Two-party Split Learning
• Regression Diagnostics meets Forecast Evaluation: Conditional Calibration, Reliability Diagrams, and Coefficient of Determination
• Uncertainty Aware Training to Improve Deep Learning Model Calibration for Classification of Cardiac MR Images
• Hyperparameter Ensembles for Robustness and Uncertainty Quantification
• Uncertainty as a Form of Transparency: Measuring, Communicating, and Using Uncertainty
• Assaying Out-Of-Distribution Generalization in Transfer Learning
• Mitigating Bias in Calibration Error Estimation
• Intra Order-preserving Functions for Calibration of Multi-Class Neural Networks
• Assessing Generalization of SGD via Disagreement
• Prior and Posterior Networks: A Survey on Evidential Deep Learning Methods For Uncertainty Estimation
• CRUDE: Calibrating Regression Uncertainty Distributions Empirically
• Multi-Class Uncertainty Calibration via Mutual Information Maximization-based Binning
• Quantifying Deep Learning Model Uncertainty in Conformal Prediction
• Calibration of Model Uncertainty for Dropout Variational Inference
• Combining Ensembles and Data Augmentation can Harm your Calibration
• Confidence-Calibrated Face and Kinship Verification
• Revisiting the Evaluation of Uncertainty Estimation and Its Application to Explore Model Complexity-Uncertainty Trade-Off
• Interpretable Uncertainty Quantification in AI for HEP
• On the use of uncertainty in classifying Aedes Albopictus mosquitoes
• Calibrate and Prune: Improving Reliability of Lottery Tickets Through Prediction Calibration
• Temporal Probability Calibration
• DICE: Diversity in Deep Ensembles via Conditional Redundancy Adversarial Estimation
• Astronomical Image Quality Prediction based on Environmental and Telescope Operating Conditions
• Improving Uncertainty-Error Correspondence in Deep Bayesian Medical Image Segmentation
• Making Heads and Tails of Models with Marginal Calibration for Sparse Tagsets
• Confidence Calibration with Bounded Error Using Transformations
• Predicting adverse outcomes following catheter ablation treatment for atrial fibrillation
• The Science of Rejection: A Research Area for Human Computation
• Scene Uncertainty and the Wellington Posterior of Deterministic Image Classifiers