Memo: Test time robustness via adaptation and augmentation

M Zhang, S Levine, C Finn - Advances in neural information …, 2022 - proceedings.neurips.cc
While deep neural networks can attain good accuracy on in-distribution test points, many
applications require robustness even in the face of unexpected perturbations in the input …

The many faces of robustness: A critical analysis of out-of-distribution generalization

D Hendrycks, S Basart, N Mu… - Proceedings of the …, 2021 - openaccess.thecvf.com
We introduce four new real-world distribution shift datasets consisting of changes in image
style, image blurriness, geographic location, camera operation, and more. With our new …

Improving robustness against common corruptions by covariate shift adaptation

S Schneider, E Rusak, L Eck… - Advances in neural …, 2020 - proceedings.neurips.cc
Today's state-of-the-art machine vision models are vulnerable to image corruptions like
blurring or compression artefacts, limiting their performance in many real-world applications …

Self-training with noisy student improves imagenet classification

Q Xie, MT Luong, E Hovy… - Proceedings of the IEEE …, 2020 - openaccess.thecvf.com
We present a simple self-training method that achieves 88.4% top-1 accuracy on ImageNet,
which is 2.0% better than the state-of-the-art model that requires 3.5 B weakly labeled …

3d common corruptions and data augmentation

OF Kar, T Yeo, A Atanov… - Proceedings of the IEEE …, 2022 - openaccess.thecvf.com
We introduce a set of image transformations that can be used as corruptions to evaluate the
robustness of models as well as data augmentation mechanisms for training neural …

Pretrained transformers improve out-of-distribution robustness

D Hendrycks, X Liu, E Wallace, A Dziedzic… - arXiv preprint arXiv …, 2020 - arxiv.org
Although pretrained Transformers such as BERT achieve high accuracy on in-distribution
examples, do they generalize to new distributions? We systematically measure out-of …

Self-supervised learning through the eyes of a child

E Orhan, V Gupta, BM Lake - Advances in Neural …, 2020 - proceedings.neurips.cc
Within months of birth, children develop meaningful expectations about the world around
them. How much of this early knowledge can be explained through generic learning …

On interaction between augmentations and corruptions in natural corruption robustness

E Mintun, A Kirillov, S Xie - Advances in Neural Information …, 2021 - proceedings.neurips.cc
Invariance to a broad array of image corruptions, such as warping, noise, or color shifts, is
an important aspect of building robust models in computer vision. Recently, several new …

Testing robustness against unforeseen adversaries

D Kang, Y Sun, D Hendrycks, T Brown, J Steinhardt - 2019 - openreview.net
Most existing defenses against adversarial attacks only consider robustness to L_p-
bounded distortions. In reality, the specific attack is rarely known in advance and …

Out-of-scope intent detection with self-supervision and discriminative training

LM Zhan, H Liang, B Liu, L Fan, XM Wu… - arXiv preprint arXiv …, 2021 - arxiv.org
Out-of-scope intent detection is of practical importance in task-oriented dialogue systems.
Since the distribution of outlier utterances is arbitrary and unknown in the training stage …