Improved DNN calibration without sacrificing accuracy.
problem Poor calibration of over-parametrized DNNs in safety-critical applications.
method Decoupling feature extraction and classification layers, and applying Gaussian priors.
result Significant improvement in model calibration with minimal training cost.
Paper presents unsupervised calibration for split conformal classification.
problem Inconvenient requirement of labeled calibration samples.
method Uses unsupervised calibration samples alongside supervised training samples.
result Achieves comparable performance to supervised calibration methods.
PosCal training improves classification models by calibrating posterior probabilities.
problem Poorly calibrated posterior probabilities in classification models.
method End-to-end training procedure that directly optimizes the objective while minimizing the difference between predicted and empirical posterior probabilities.
result PosCal training achieves about 2.5% task performance gain and 16.1% calibration error reduction.
Unified calibration metrics improve forecast sharpness and accuracy.
problem Improving the sharpness of probabilistic forecasts while maintaining calibration.
method Kernel-based calibration metrics that unify and generalize existing methods for classification and regression.
result Enhanced calibration, sharpness, and decision-making across various tasks.
In safety-critical applications a probabilistic model is usually required to be calibrated, i.e., to capture the uncertainty of its predictions accurately. In multi-class classification, calibration of the most confident predictions only is often not sufficient. We propose and study calibration measures for multi-class…
JUCAL jointly calibrates aleatoric and epistemic uncertainties in classifier ensembles.
problem Misrepresentation of predictive uncertainty due to unbalanced aleatoric and epistemic uncertainties.
method Joint Uncertainty Calibration (JUCAL) that jointly calibrates two constants to weight and scale uncertainties.
result Significantly outperforms state-of-the-art calibration methods across various text classification tasks.
Paper proposes a unified framework for evaluating calibration of probabilistic models.
problem Evaluation of calibration for general probabilistic predictive models.
method Unified framework for calibration evaluation and tests for any probabilistic model.
result Generalization and reformulation of existing measures and tests.
The paper reviews and extends calibration concepts for classification and regression.
problem Formalizing compatibility between probabilistic predictions and outcomes.
method Review and extension of existing calibration concepts, introduction of new concepts.
result Hierarchical relations between calibration concepts for various data types.
Optimizes calibration error estimators for better classifier trustworthiness.
problem Lack of guidance on selecting and tuning calibration error estimators.
method Reformulates calibration estimation as a regression problem with i.i.d. input pairs.
result Demonstrates the effectiveness of optimized calibration estimators on image classification tasks.
A method to monitor probability predictions for calibration loss in image classification models.
problem Maintaining calibration in machine learning predictions over time.
method Cumulative sum-based approach with dynamic limits for early detection of miscalibration.
result Early detection of operational context changes impacting image classification performance.
Establishes a condition for multiclass classification-calibration of Gamma-Phi losses.
problem Ensuring classification-calibration of multiclass Gamma-Phi losses.
method Develops a general sufficient condition for classification-calibration of Gamma-Phi losses.
result Proves the first family of nonconvex multiclass surrogate losses for which classification-calibration has been fully justified.
Survey on assessing and improving classifier calibration for better decision making.
problem Ensuring classifiers correctly quantify prediction uncertainty.
method Overview of principles, methods, and evaluation metrics for calibration.
result New methods and extensions from binary to multiclass settings.
Study compares various calibration methods for binary classification tasks.
problem Improving probabilistic predictions in binary classification models.
method Benchmarked 21 classifiers using 5 calibration methods on real data.
result Venn-Abers predictors and Beta calibration show the largest log-loss reductions.
Study three types of uncertainty quantification for binary classification without distributional assumptions.
problem Uncertainty quantification for binary classification in a distribution-free setting.
method Established theorems connecting calibration, confidence intervals, and prediction sets for score-based classifiers.
result Distribution-free calibration is only possible using scoring functions that partition feature space into countably many sets.
Improves model calibration for deep neural networks using proper scores.
problem Calibration errors in deep neural networks are often biased and inconsistent.
method Introduces proper calibration errors related to proper scores.
result Demonstrates the superiority of proper scores over common estimators.
Proposes MCLLO for assessing and recalibrating multiclass probability predictions.
problem Limited multicategory recalibration methods for assessing and comparing model calibration.
method MCLLO recalibration method that assesses calibration without model access and is easy to interpret.
result MCLLO outperforms other methods in simulations and real-world case studies.
Study on estimating conditional risk in machine learning.
problem Estimating expected loss of prediction models given input features.
method Analyzed in classification and regression settings, showing equivalence to standard regression. Developed theoretical insights and empirical validation.
result Conditional risk calibration is distinct from existing uncertainty quantification problems.
Proposes a new method for multi-class classification with well-calibrated predictions.
problem Improving the accuracy and reliability of multi-class classification models.
method Trains data in a latent space induced by an (n−1)-dimensional simplex, then extends and fits a regression model. result Demonstrates a well-calibrated classifier with improved prediction and calibration properties.
Develops geometric framework for uncertainty-aware multi-class classification.
problem Silent failure of AI models when uncertain, especially in multi-class settings.
method Geometric framework treating probability vectors as points on the (c−1)-dimensional probability simplex, using Fisher--Rao metric for calibration and uncertainty quantification. result Empirical validation shows 72.5% of errors captured while deferring 34.5% of ambiguous predictions, reducing automated decision error rates from 16.8% to 6.9%.
Framework calibrates ML models for risk control in various tasks.
problem Achieving statistical guarantees for model predictions.
method Reframing risk control as multiple hypothesis testing, applying statistical techniques.
result New calibration methods for multi-label classification, instance segmentation, outlier detection, and confidence set coverage.
New method calibrates classifier probabilities with guaranteed coverage.
problem Inaccurate probability estimates by classifiers in high-risk applications.
method Adaptive temperature scaling algorithm for conformal prediction.
result Improves calibration error measures and standard metrics across various tasks.
In this paper we refine the process of computing calibration functions for a number of multiclass classification surrogate losses. Calibration functions are a powerful tool for easily converting bounds for the surrogate risk (which can be computed through well-known methods) into bounds for the true risk, the probabili…
The paper introduces a spline-based method for calibrating neural networks.
problem Ensuring neural network outputs are reliable for safety-critical applications.
method Approximating the empirical cumulative distribution function using splines to map network outputs to calibrated probabilities.
result The spline-based recalibration consistently outperforms existing methods on calibration measures.
New method calibrates machine learning models with theoretical guarantees.
problem Lack of theoretical guarantees for recalibration in multiclass classification.
method PAC-Bayes analysis for generalization error in calibration.
result First optimizable upper bound for generalization error in calibration.
Proposes an accuracy-preserving calibration method for DNNs.
problem Calibration of deep neural networks (DNNs) to measure prediction reliability.
method Uses Concrete distribution on the probability simplex to calibrate DNNs without accuracy loss.
result The proposed method outperforms previous methods in accuracy-preserving calibration tasks.
We present surrogate regret bounds for arbitrary surrogate losses in the context of binary classification with label-dependent costs. Such bounds relate a classifier's risk, assessed with respect to a surrogate loss, to its cost-sensitive classification risk. Two approaches to surrogate regret bounds are developed. The…
Nested dichotomies are used as a method of transforming a multiclass classification problem into a series of binary problems. A tree structure is induced that recursively splits the set of classes into subsets, and a binary classification model learns to discriminate between the two subsets of classes at each node. In …
CRUDE calibrates regression uncertainty without assuming specific error distributions.
problem Uncalibrated uncertainty estimates in regression models, especially for modern predictive tasks.
method CRUDE assumes error distributions have a constant shape, shifted and scaled by predicted mean and standard deviation.
result CRUDE produces sharper, better calibrated, and more accurate uncertainty estimates than existing methods.
Tree-based models biased when trained on imbalanced data, requiring new calibration methods.
problem Bias in tree-based models trained on imbalanced datasets.
method Analytical calibration of random forest models, demonstrating bias in decision trees.
result Calibrating tree-based models on imbalanced data negatively impacts predictions, especially for the minority class.
Unified binary and multiclass margin-based classification methods.
problem No consensus on multiclass loss functions analogous to binary margin loss.
method Showed multiclass loss functions can be expressed in relative margin form.
result Extended classification-calibration result to multiclass.
Many applications of classification methods not only require high accuracy but also reliable estimation of predictive uncertainty. However, while many current classification frameworks, in particular deep neural networks, achieve high accuracy, they tend to incorrectly estimate uncertainty. In this paper, we propose a …
The paper tackles confidence calibration for exploratory machine learning problems.
problem Difficulty in curating datasets and confusion about category validity.
method Introduces four new algorithms for category-specific confidence estimation, including kernel density ratios.
result Kernel density ratios provide a novel approach to confidence calibration, especially for exploratory problems.
Study calibrates high-dimensional binary classifiers using angle between estimator and true weights.
problem Calibrating high-dimensional binary classifiers with provable properties.
method Interpolates with a chance classifier to construct well-calibrated predictor based on angle between estimator and true weights.
result Angular calibration approach is provably well-calibrated in high dimensions, minimizing Bregman divergence.
Excellent ranking power along with well calibrated probability estimates are needed in many classification tasks. In this paper, we introduce a technique, Calibrated Boosting-Forest that captures both. This novel technique is an ensemble of gradient boosting machines that can support both continuous and binary labels. …
Deep neural networks (DNNs) are poorly calibrated when trained in conventional ways. To improve confidence calibration of DNNs, we propose a novel training method, distance-based learning from errors (DBLE). DBLE bases its confidence estimation on distances in the representation space. In DBLE, we first adapt prototypi…
CalArena benchmarks post-hoc calibration methods across various tasks.
problem Inconsistent evaluations of post-hoc calibration methods.
method Large-scale benchmark with 2000 experiments, covering diverse models and settings.
result Smooth calibration functions outperform binning-based approaches.
This paper characterizes and designs loss functions for robust classification with abstention.
problem Ensuring robustness against adversarial attacks and knowing when to abstain from prediction.
method Proposes adversarial robust reject option loss and characterizes surrogates for calibration.
result Shifted Double Ramp Loss and Shifted Double Sigmoid Loss satisfy the calibration conditions.
This research analyzes the consistency of convex and nonconvex surrogate losses for adversarially robust classification.
problem Ensuring classifiers are robust to adversarial perturbations.
method Analysis of convex and nonconvex surrogate losses through the lens of calibration.
result No convex surrogate loss is calibrated with respect to the adversarial 0-1 loss for linear models, but nonconvex losses can be calibrated under certain conditions.
Mixup technique improved, reducing manifold mismatch for better calibration.
problem Improving calibration of models using Mixup.
method Dynamic adjustment of interpolation coefficients based on sample similarity.
result Improved predictive performance and calibration with reduced manifold mismatch.
Binary classification is highly used in credit scoring in the estimation of probability of default. The validation of such predictive models is based both on rank ability, and also on calibration (i.e. how accurately the probabilities output by the model map to the observed probabilities). In this study we cover the cu…
New PAC-Bayes meta-learning method improves few-shot learning accuracy and calibration.
problem Few-shot learning with limited data.
method PAC-Bayes framework extended to meta-learning, estimating task-specific posteriors.
result State-of-the-art calibration and classification results on benchmarks.
New method calibrates deep models for both in-distribution and out-of-distribution samples.
problem Ensuring calibration for deep models in safety-critical applications, especially in OOD regions.
method Geodesic distance and Gaussian kernel to calibrate deep models.
result Proposed KDF and KDN methods achieve well-calibrated posteriors for both in-distribution and out-of-distribution samples.
Advances rule-based multi-label classification using conformal prediction.
problem Improving accuracy and decision making in multi-label classification.
method Combines conformal prediction with rule-based learning to provide natural conformity scores and calibrate rule assessments.
result Calibrated conformity scores enhance prediction accuracy and decision making.
We improve adversarial robustness calibration analysis for broader hypothesis sets.
problem Improving calibration for adversarial robustness in machine learning.
method A finer definition of calibration for adversarial robustness.
result Our results cover most common hypothesis sets in machine learning.
Obtaining accurate and well calibrated probability estimates from classifiers is useful in many applications, for example, when minimising the expected cost of classifications. Existing methods of calibrating probability estimates are applied globally, ignoring the potential for improvements by applying a more fine-gra…
Loss-calibrated EP improves Bayesian decision-making by focusing on utility-sensitive posterior approximations.
problem Bayesian decision-making under asymmetric utility functions.
method Loss-calibrated expectation propagation (Loss-EP) that tilts the posterior towards higher utility decisions.
result Loss-EP can capture useful information for decision-making under asymmetric penalties.
A new framework separates classifier calibration and discrimination.
problem Combining reliability and resolution in probabilistic predictions.
method Manokhin Probability Matrix separates reliability and resolution using Spiegelhalter Z-statistic and AUC-ROC.
result Classifiers are categorized into four archetypes: Eagle, Bull, Sloth, and Mole.
A post-hoc framework improves model performance by calibrating different feature spaces.
problem Improving AUC performance on binary classification tasks for overconfident models.
method Identifies heterogeneous partitions of the feature space and applies post-hoc calibration techniques to each partition.
result Theoretical optimality of the framework for any model, demonstrated on deep neural networks.