Out-of-Domain Robustness via Targeted Augmentations
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2023
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866917583184199680 |
|---|---|
| author | Gao, Irena Sagawa, Shiori Koh, Pang Wei Hashimoto, Tatsunori Liang, Percy |
| author_facet | Gao, Irena Sagawa, Shiori Koh, Pang Wei Hashimoto, Tatsunori Liang, Percy |
| contents | Models trained on one set of domains often suffer performance drops on unseen domains, e.g., when wildlife monitoring models are deployed in new camera locations. In this work, we study principles for designing data augmentations for out-of-domain (OOD) generalization. In particular, we focus on real-world scenarios in which some domain-dependent features are robust, i.e., some features that vary across domains are predictive OOD. For example, in the wildlife monitoring application above, image backgrounds vary across camera locations but indicate habitat type, which helps predict the species of photographed animals. Motivated by theoretical analysis on a linear setting, we propose targeted augmentations, which selectively randomize spurious domain-dependent features while preserving robust ones. We prove that targeted augmentations improve OOD performance, allowing models to generalize better with fewer domains. In contrast, existing approaches such as generic augmentations, which fail to randomize domain-dependent features, and domain-invariant augmentations, which randomize all domain-dependent features, both perform poorly OOD. In experiments on three real-world datasets, we show that targeted augmentations set new states-of-the-art for OOD performance by 3.2-15.2 percentage points. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2302_11861 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Out-of-Domain Robustness via Targeted Augmentations Gao, Irena Sagawa, Shiori Koh, Pang Wei Hashimoto, Tatsunori Liang, Percy Machine Learning Computer Vision and Pattern Recognition Models trained on one set of domains often suffer performance drops on unseen domains, e.g., when wildlife monitoring models are deployed in new camera locations. In this work, we study principles for designing data augmentations for out-of-domain (OOD) generalization. In particular, we focus on real-world scenarios in which some domain-dependent features are robust, i.e., some features that vary across domains are predictive OOD. For example, in the wildlife monitoring application above, image backgrounds vary across camera locations but indicate habitat type, which helps predict the species of photographed animals. Motivated by theoretical analysis on a linear setting, we propose targeted augmentations, which selectively randomize spurious domain-dependent features while preserving robust ones. We prove that targeted augmentations improve OOD performance, allowing models to generalize better with fewer domains. In contrast, existing approaches such as generic augmentations, which fail to randomize domain-dependent features, and domain-invariant augmentations, which randomize all domain-dependent features, both perform poorly OOD. In experiments on three real-world datasets, we show that targeted augmentations set new states-of-the-art for OOD performance by 3.2-15.2 percentage points. |
| title | Out-of-Domain Robustness via Targeted Augmentations |
| topic | Machine Learning Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2302.11861 |