BREEDS: Benchmarks for Subpopulation Shift

By Shibani Santurkar, Dimitris Tsipras, Aleksander Madry. 2020

195 citations recorded by the source corpus.

We develop a methodology for assessing the robustness of models to subpopulation shift---specifically, their ability to generalize to novel data subpopulations that were not observed during training. Our approach leverages the class structure underlying existing datasets to control the data subpopulations that comprise the training and test distributions. This enables us to synthesize realistic distribution shifts whose sources can be precisely controlled and characterized, within existing large-scale datasets. Applying this methodology to the ImageNet dataset, we create a suite of subpopulation shift benchmarks of varying granularity. We then validate that the corresponding shifts are tractable by obtaining human baselines for them. Finally, we utilize these benchmarks to measure the sensitivity of standard model architectures as well as the effectiveness of off-the-shelf train-time robustness interventions. Code and data available at this https URL .

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

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 Computer Vision and Pattern Recognition

Citations in the corpus

IIRC: Incremental Implicitly-Refined Classification (2020, background): “Recently, BREEDS [41] proposed a benchmark to evaluate model’s generalization capabilities in the context of subpopulation shift.”
WILDS: A Benchmark of in-the-Wild Distribution Shifts (2020, neutral): “…Remixed (Hendrycks et al., 2020), which involves classifying objects at different scales solely from objects at a single scale; or BREEDS (Santurkar et al., 2020) and a similar dataset (Hendrycks and Dietterich, 2019), which use ImageNet training sets with disjoint subclasses from the…”
Calibrated ensembles can mitigate accuracy tradeoffs under distribution shift (2022, background): “Consider Breeds Living-17 [Santurkar et al., 2020] where the goal is to classify an image as one of 17 animal categories.”
In-N-Out: Pre-Training and Self-Training using Auxiliary Information for Out-of-Distribution Robustness (2020, result): “We plot the in-distribution and OOD accuracy for each model, showing that there is a significant correlation between in-distribution and OOD accuracy in CelebA, supporting results on standard datasets [37, 53, 41].”
Environment Inference for Invariant Learning (2020, background): “, 2021), handling subpopulation shift (Santurkar et al., 2021), and model selection (Gulrajani & Lopez-Paz, 2021).”
SmoothLLM: Defending Large Language Models Against Jailbreaking Attacks (2023, unknown): “Such changes—include nuisances in visual data [49–51], sub-population drift [52, 53], and distribution shift [54–57]—limit the applicability of deep learning methods in safety critical areas.”
Balancing Between Forgetting and Acquisition in Incremental Subpopulation Learning (2022, methodology): “Santurkar et al. [28] studied how well a model generalizes to subpopulations that are unseen during training, i.e., whether the model can recognize “Dalmatians” as “dogs” even their training data for “dogs” comprise only the dog’s breeds like “Poodles” and “Terriers”.”
Distilling Model Failures as Directions in Latent Space (2022, unknown): “…on speciﬁc subpopulations in the data if that subpopulation was mislabeled (Northcutt et al., 2021; Stock & Cisse, 2018; Beyer et al., 2020; Vasudevan et al., 2022), underrepresented (Sagawa et al., 2020; Santurkar et al., 2021), or corrupted (Hendrycks & Dietterich, 2019; Hendrycks et al., 2020).”
TRAK: Attributing Model Behavior at Scale (2023, unknown): “The LIVING17 dataset (Santurkar et al., 2021) is an image classification dataset derived from the ImageNet dataset and consists of 17 classes, each comprised of four original ImageNet classes.”