On Deep Neural Network Calibration by Regularization and its Impact on Refinement
Not on the map. Why?
Abstract (arXiv)
Deep neural networks have been shown to be highly miscalibrated. often they tend to be overconfident in their predictions. It poses a significant challenge for safety-critical systems to utilise deep neural networks (DNNs), reliably. Many recently proposed approaches to mitigate this have demonstrated substantial progress in improving DNN calibration. However, they hardly touch upon refinement, which historically has been an essential aspect of calibration. Refinement indicates separability of a network's correct and incorrect predictions. This paper presents a theoretically and empirically supported exposition reviewing refinement of a calibrated model. Firstly, we show the breakdown of expected calibration error (ECE), into predicted confidence and refinement under the assumption of over-confident predictions. Secondly, linking with this result, we highlight that regularization based calibration only focuses on naively reducing a model's confidence. This logically has a severe downside to a model's refinement as correct and incorrect predictions become tightly coupled. Lastly, connecting refinement with ECE also provides support to existing refinement based approaches which improve calibration but do not explain the reasoning behind it. We support our claims through rigorous empirical evaluations of many state of the art calibration approaches on widely used datasets and neural networks. We find that many calibration approaches with the likes of label smoothing, mixup etc. lower the usefulness of a DNN by degrading its refinement. Even under natural data shift, this calibration-refinement trade-off holds for the majority of calibration methods.
Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Builds on 30 of its 62 references that are in the map. First posted on arXiv 17 Jun 2021.
Builds on
30 of 62 references.
References · 30
- 2015Deep Residual Learning for Image Recognitioncited by 51,809
- 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,160
- 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,267
- 2016Densely Connected Convolutional Networkscited by 6,691
- 2017mixup: Beyond Empirical Risk Minimizationcited by 5,155
- 2017On Calibration of Modern Neural Networkscited by 4,860
- 2016Simple and Scalable Predictive Uncertainty Estimation using Deep Ensemblescited by 3,888
- 2016A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networkscited by 2,362
- 2019Do ImageNet Classifiers Generalize to ImageNet?cited by 1,507
- 2019Can You Trust Your Model's Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shiftcited by 1,274
- 2019AugMix: A Simple Data Processing Method to Improve Robustness and Uncertaintycited by 897
- 2019Using Self-Supervised Learning Can Improve Model Robustness and Uncertaintycited by 511
18 more
- 2017Regularizing Neural Networks by Penalizing Confident Output Distributionscited by 461
- 2020Measuring Robustness to Natural Distribution Shifts in Image Classificationcited by 452
- 2019Measuring Calibration in Deep Learningcited by 443
- 2019Using Pre-Training Can Improve Model Robustness and Uncertaintycited by 373
- 2019Beyond temperature scaling: Obtaining well-calibrated multiclass probabilities with Dirichlet calibrationcited by 359
- 2020Calibrating Deep Neural Networks using Focal Losscited by 353
- 2019Verified Uncertainty Calibrationcited by 307
- 2019On Mixup Training: Improved Calibration and Predictive Uncertainty for Deep Neural Networkscited by 304
- 2021Escaping the Big Data Paradigm with Compact Transformerscited by 261
- 2019Human uncertainty makes classification more robustcited by 231
- 2018To Trust Or Not To Trust A Classifiercited by 205
- 2018Do CIFAR-10 Classifiers Generalize to CIFAR-10?cited by 197
- 2019Addressing Failure Prediction by Learning Model Confidencecited by 183
- 2020Confidence-Aware Learning for Deep Neural Networkscited by 113
- 2018Bias-Reduced Uncertainty Estimation for Deep Neural Classifierscited by 106
- 2019Calibration tests in multi-class classification: A unifying frameworkcited by 79
- 2021Don't Just Blame Over-parametrization for Over-confidence: Theoretical Analysis of Calibration in Binary Classificationcited by 39
- 2021Meta-Cal: Well-controlled Post-hoc Calibration by Rankingcited by 33
Built on by
Citing papers per year: 2022: 1; 2023: 0; 2024: 0; 2025: 1.
2 papers.