Human Uncertainty Makes Classification More Robust

By Joshua C. Peterson, Ruairidh M. Battleday, T. Griffiths, Olga Russakovsky. 2019

386 citations recorded by the source corpus.

The classification performance of deep neural networks has begun to asymptote at near-perfect levels. However, their ability to generalize outside the training set and their robustness to adversarial attacks have not. In this paper, we make progress on this problem by training with full label distributions that reflect human perceptual uncertainty. We first present a new benchmark dataset which we call CIFAR10H, containing a full distribution of human labels for each image of the CIFAR10 test set. We then show that, while contemporary classifiers fail to exhibit human-like uncertainty on their own, explicit training on our dataset closes this gap, supports improved generalization to increasingly out-of-training-distribution test datasets, and confers robustness to adversarial attacks.

Subfields: Artificial Intelligence, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Safety, Robustness & Interpretability

Citations in the corpus

Are We Done with MMLU? (2024, neutral): “…[28, 29, 30]; automated labelling processes such as crowdsourcing [31] — where annotators may be optimizing for their own incentives; human errors [32] — e.g., lack of expertise on a given topic; or programmatic weak supervision [33, 34, 7], and personal [35] or group-level [36] annotator biases.”
Consistent Estimators for Learning to Defer to an Expert (2020, methodology): “To experiment in settings where we have limited expert data, we use the dataset CIFAR10H (Peterson et al., 2019) initially developed to improve model robustness.”
Learning with Noisy Labels Revisited: A Study Using Real-World Human Annotations (2021, background): “CIFAR10H [Peterson et al., 2019] only provides test images with noisy human annotations.”
FMix: Enhancing Mixed Sample Data Augmentation (2020, methodology): “Peterson et al. (2019) suggest using human uncertainty as a baseline on the CIFAR-10 data set.”
Pitfalls in Machine Learning Research: Reexamining the Development Cycle (2020, neutral): “Rater reliability can be assessed [41, 35], but rarely is.”
Self-Distillation as Instance-Specific Label Smoothing (2020, result): “Previous results show that training with such instancespecific uncertainty can indeed lead to better performance [30].”
A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges (2020, background): “[10], optical flow estimation and motion prediction [249], [410], [411], human pose estimation and pedestrian localization [412], [413], [305], person re-identification and face recognition [414], [415], [416], camera re-localization [397], avoiding adversarial attacks [417], [418], during the years”
Delving Deep Into Label Smoothing (2020, neutral): “For PGD, we apply the same experimental setup as in [72] except that we increase the iteration times to 20, which is enough to get better attack effects.”