Training image classifiers on human label distributions rather than one-hot labels improves out-of-distribution generalization and adversarial robustness, as demonstrated with the new CIFAR10H dataset.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Human uncertainty makes classification more robust
Training image classifiers on human label distributions rather than one-hot labels improves out-of-distribution generalization and adversarial robustness, as demonstrated with the new CIFAR10H dataset.