Label shift refers to the phenomenon where the marginal probability p(y) of
observing a particular class changes between the training and test
distributions while the conditional probability p(x|y) stays fixed. This is
relevant in settings such as medical diagnosis, where a classifier trained to
predict disease based on observed symptoms may need to be adapted to a
different distribution where the baseline frequency of the disease is higher.
Given calibrated estimates of p(y|x), one can apply an EM algorithm to correct
for the shift in class imbalance between the training and test distributions
without ever needing to calculate p(x|y). Unfortunately, modern neural networks
typically fail to produce well-calibrated probabilities, compromising the
effectiveness of this approach. Although Temperature Scaling can greatly reduce
miscalibration in these networks, it can leave behind a systematic bias in the
probabilities that still poses a problem. To address this, we extend
Temperature Scaling with class-specific bias parameters, which largely
eliminates systematic bias in the calibrated probabilities and allows for
effective domain adaptation under label shift. We term our calibration approach
"Bias-Corrected Temperature Scaling". On experiments with CIFAR10, we find that
EM with Bias-Corrected Temperature Scaling significantly outperforms both EM
with Temperature Scaling and the recently-proposed Black-Box Shift Estimation.
Description
[1901.06852] Calibration with Bias-Corrected Temperature Scaling Improves Domain Adaptation Under Label Shift in Modern Neural Networks
%0 Journal Article
%1 shrikumar2019calibration
%A Shrikumar, Avanti
%A Kundaje, Anshul
%D 2019
%K calibration uncertainty
%T Calibration with Bias-Corrected Temperature Scaling Improves Domain
Adaptation Under Label Shift in Modern Neural Networks
%U http://arxiv.org/abs/1901.06852
%X Label shift refers to the phenomenon where the marginal probability p(y) of
observing a particular class changes between the training and test
distributions while the conditional probability p(x|y) stays fixed. This is
relevant in settings such as medical diagnosis, where a classifier trained to
predict disease based on observed symptoms may need to be adapted to a
different distribution where the baseline frequency of the disease is higher.
Given calibrated estimates of p(y|x), one can apply an EM algorithm to correct
for the shift in class imbalance between the training and test distributions
without ever needing to calculate p(x|y). Unfortunately, modern neural networks
typically fail to produce well-calibrated probabilities, compromising the
effectiveness of this approach. Although Temperature Scaling can greatly reduce
miscalibration in these networks, it can leave behind a systematic bias in the
probabilities that still poses a problem. To address this, we extend
Temperature Scaling with class-specific bias parameters, which largely
eliminates systematic bias in the calibrated probabilities and allows for
effective domain adaptation under label shift. We term our calibration approach
"Bias-Corrected Temperature Scaling". On experiments with CIFAR10, we find that
EM with Bias-Corrected Temperature Scaling significantly outperforms both EM
with Temperature Scaling and the recently-proposed Black-Box Shift Estimation.
@article{shrikumar2019calibration,
abstract = {Label shift refers to the phenomenon where the marginal probability p(y) of
observing a particular class changes between the training and test
distributions while the conditional probability p(x|y) stays fixed. This is
relevant in settings such as medical diagnosis, where a classifier trained to
predict disease based on observed symptoms may need to be adapted to a
different distribution where the baseline frequency of the disease is higher.
Given calibrated estimates of p(y|x), one can apply an EM algorithm to correct
for the shift in class imbalance between the training and test distributions
without ever needing to calculate p(x|y). Unfortunately, modern neural networks
typically fail to produce well-calibrated probabilities, compromising the
effectiveness of this approach. Although Temperature Scaling can greatly reduce
miscalibration in these networks, it can leave behind a systematic bias in the
probabilities that still poses a problem. To address this, we extend
Temperature Scaling with class-specific bias parameters, which largely
eliminates systematic bias in the calibrated probabilities and allows for
effective domain adaptation under label shift. We term our calibration approach
"Bias-Corrected Temperature Scaling". On experiments with CIFAR10, we find that
EM with Bias-Corrected Temperature Scaling significantly outperforms both EM
with Temperature Scaling and the recently-proposed Black-Box Shift Estimation.},
added-at = {2019-08-22T20:39:30.000+0200},
author = {Shrikumar, Avanti and Kundaje, Anshul},
biburl = {https://www.bibsonomy.org/bibtex/2a9dc10d4fa04337c637b78beee2f4518/kirk86},
description = {[1901.06852] Calibration with Bias-Corrected Temperature Scaling Improves Domain Adaptation Under Label Shift in Modern Neural Networks},
interhash = {0cd9f4527b49dec603e88c5806614b60},
intrahash = {a9dc10d4fa04337c637b78beee2f4518},
keywords = {calibration uncertainty},
note = {cite arxiv:1901.06852},
timestamp = {2019-08-22T20:39:30.000+0200},
title = {Calibration with Bias-Corrected Temperature Scaling Improves Domain
Adaptation Under Label Shift in Modern Neural Networks},
url = {http://arxiv.org/abs/1901.06852},
year = 2019
}