Hidden Heterogeneity: When to Choose Similarity-Based Calibration (2022-02-03T00:00:00.000000Z)

TL;DR

HH can serve as a useful diagnostic tool for identifying when local calibration methods would be beneficial, and two similarity-weighted calibration methods that can address HH by adapting locally to each test item are introduced.

Abstract

Trustworthy classifiers are essential to the adoption of machine learning predictions in many real-world settings. The predicted probability of possible outcomes can inform high-stakes decision making, particularly when assessing the expected value of alternative decisions or the risk of bad outcomes. These decisions require well-calibrated probabilities, not just the correct prediction of the most likely class. Black-box classifier calibration methods can improve the reliability of a classifier's output without requiring retraining. However, these methods are unable to detect subpopulations where calibration could also improve prediction accuracy. Such subpopulations are said to exhibit"hidden heterogeneity"(HH), because the original classifier did not detect them. This paper proposes a quantitative measure for HH. It also introduces two similarity-weighted calibration methods that can address HH by adapting locally to each test item: SWC weights the calibration set by similarity to the test item, and SWC-HH explicitly incorporates hidden heterogeneity to filter the calibration set. Experiments show that the improvements in calibration achieved by similarity-based calibration methods correlate with the amount of HH present and, given sufficient calibration data, generally exceed calibration achieved by global methods. HH can therefore serve as a useful diagnostic tool for identifying when local calibration methods would be beneficial.

Authors

Thomas G. Dietterich

6 papers

K. Wagstaff

1 papers

TL;DR

Abstract

Authors

References49 items

Oracle Analysis of Representations for Deep Open Set Detection

Robust Calibration with Multi-domain Temperature Scaling

The Familiarity Hypothesis: Explaining the Behavior of Deep Open Set Methods

Revisiting the Calibration of Modern Neural Networks

Learning to Cascade: Confidence Calibration for Improving the Accuracy and Computational Cost of Cascade Inference Systems

Confidence Calibration for Domain Generalization under Covariate Shift

RepVGG: Making VGG-style ConvNets Great Again

An Empirical Analysis of Backward Compatibility in Machine Learning Systems

Multi-Class Uncertainty Calibration via Mutual Information Maximization-based Binning

Individual Calibration with Randomized Forecasting

Mix-n-Match: Ensemble and Compositional Methods for Uncertainty Calibration in Deep Learning

Calibrating Deep Neural Networks using Focal Loss

Calibration tests in multi-class classification: A unifying framework

Towards Trustworthy Predictions from Deep Neural Networks with Fast Adversarial Calibration

Verified Uncertainty Calibration

Beyond temperature scaling: Obtaining well-calibrated multiclass probabilities with Dirichlet calibration

Can You Trust Your Model's Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shift

Measuring Calibration in Deep Learning

Evaluating model calibration in classification

Learning for Single-Shot Confidence Calibration in Deep Neural Networks Through Stochastic Inferences

Trainable Calibration Measures For Neural Networks From Kernel Mean Embeddings

Beyond sigmoids: How to obtain well-calibrated probabilities from binary classifiers with beta calibration

Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms

On Calibration of Modern Neural Networks

Deep Residual Learning for Image Recognition

Random Forests and Kernel Methods

Obtaining Well Calibrated Probabilities Using Bayesian Binning

Efficient Nearest-Neighbor Search in the Probability Simplex

Estimating reliability and resolution of probability forecasts through decomposition of the empirical score

A bias‐corrected decomposition of the Brier score

On the effect of calibration in classifier combination

Classification and regression trees

Scikit-learn: Machine Learning in Python

Multiple Kernel Learning Algorithms

Similarity-Binning Averaging: A Generalisation of Binning Calibration

Predicting good probabilities with supervised learning

The Elements of Statistical Learning: Data Mining, Inference, and Prediction

Transforming classifier scores into accurate multiclass probability estimates

Obtaining calibrated probability estimates from decision trees and naive Bayesian classifiers

Stacked generalization

Localized Calibration: Metrics and Recalibration

Improving the Calibration of Long Term Predictions of Heart Failure Rehospitalizations using Medical Concept Embedding

Maximum Likelihood with Bias-Corrected Calibration is Hard-To-Beat at Label Shift Adaptation

Learning Multiple Layers of Features from Tiny Images

Letter recognition using Holland-style adaptive classifiers

Random Forests

Probabilistic Outputs for Support vector Machines and Comparisons to Regularized Likelihood Methods

Gradient-based learning applied to document recognition

VERIFICATION OF FORECASTS EXPRESSED IN TERMS OF PROBABILITY

Field of Study

Journal Information

Name

Volume