Corpus ID: 203593350

Well-calibrated Model Uncertainty with Temperature Scaling for Dropout Variational Inference

@article{Laves2019WellcalibratedMU,
  title={Well-calibrated Model Uncertainty with Temperature Scaling for Dropout Variational Inference},
  author={Max-Heinrich Laves and Sontje Ihler and Karl-Philipp Kortmann and T. Ortmaier},
  journal={ArXiv},
  year={2019},
  volume={abs/1909.13550}
}
Model uncertainty obtained by variational Bayesian inference with Monte Carlo dropout is prone to miscalibration. The uncertainty does not represent the model error well. In this paper, temperature scaling is extended to dropout variational inference to calibrate model uncertainty. Expected uncertainty calibration error (UCE) is presented as a metric to measure miscalibration of uncertainty. The effectiveness of this approach is evaluated on CIFAR-10/100 for recent CNN architectures… Expand
10 Citations
Improving model calibration with accuracy versus uncertainty optimization
  • 2
  • Highly Influenced
  • PDF
Well-Calibrated Regression Uncertainty in Medical Imaging with Deep Learning
  • 6
  • PDF
Multi-Loss Sub-Ensembles for Accurate Classification with Uncertainty Estimation
  • Highly Influenced
  • PDF
Uncertainty-Aware Learning for Zero-Shot Semantic Segmentation
  • 1
  • PDF
Uncertainty Intervals for Graph-based Spatio-Temporal Traffic Prediction
  • 1
  • PDF
Spatially Varying Label Smoothing: Capturing Uncertainty from Expert Annotations
  • PDF
Know Your Limits: Monotonicity & Softmax Make Neural Classifiers Overconfident on OOD Data

References

SHOWING 1-10 OF 23 REFERENCES
Concrete Dropout
  • 239
  • PDF
What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?
  • 1,390
  • Highly Influential
  • PDF
Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning
  • 2,825
  • PDF
On Calibration of Modern Neural Networks
  • 1,206
  • PDF
Auto-Encoding Variational Bayes
  • 11,285
  • PDF
Regularizing Neural Networks by Penalizing Confident Output Distributions
  • 505
  • PDF
Decoupled Weight Decay Regularization
  • 887
Obtaining Well Calibrated Probabilities Using Bayesian Binning
  • 264
  • PDF
Predicting good probabilities with supervised learning
  • 770
  • PDF
Categorical Reparameterization with Gumbel-Softmax
  • 1,703
  • PDF
...
1
2
3
...