When Robustness Doesn’t Promote Robustness: Synthetic vs. Natural Distribution Shifts on ImageNet

By Rohan Taori, Achal Dave, Vaishaal Shankar, Nicholas Carlini, B. Recht, Ludwig Schmidt. 2019

21 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, 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 Computer Vision and Pattern Recognition

1. Luc Van Gool
2. Dacheng Tao
3. Trevor Darrell
4. Ting-Li Chen
5. Andrew Zisserman
6. Ming-Hsuan Yang
7. Jitendra Malik
8. Xiaogang Wang

In Safety, Robustness & Interpretability

1. Ting-Li Chen
2. Katherine L. Hermann
3. Behnam Neyshabur
4. Rebecca Roelofs
5. Simon Kornblith

Citations in the corpus

• Measuring Robustness to Natural Distribution Shifts in Image Classification (2020, neutral): “An early version of this paper with results on ImageNetV2 and ImageNet-Vid-Robust appeared on OpenReview in late 2019 [90].”
• The Evolution of Out-of-Distribution Robustness Throughout Fine-Tuning (2021, methodology): “Models which lie above the linear fit are said to exhibit effective robustness (ER) (Taori et al., 2020), which measures the model’s OOD accuracy relative to the fit (see Figure 1(a) for an illustration).”
• The Origins and Prevalence of Texture Bias in Convolutional Neural Networks (2019, background): “However, most of these studies investigate robustness to simple image corruptions, which may not be correlated with robustness to natural distribution shift [88].”