Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2505.15191 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909942264365056 |
|---|---|
| author | Satou, Hana Mitkiy, Alan Collins, Emma Kingston, Finn |
| author_facet | Satou, Hana Mitkiy, Alan Collins, Emma Kingston, Finn |
| contents | Transfer learning under domain shift remains a fundamental challenge due to the divergence between source and target data manifolds. In this paper, we propose MAADA (Manifold-Aware Adversarial Data Augmentation), a novel framework that decomposes adversarial perturbations into on-manifold and off-manifold components to simultaneously capture semantic variation and model brittleness. We theoretically demonstrate that enforcing on-manifold consistency reduces hypothesis complexity and improves generalization, while off-manifold regularization smooths decision boundaries in low-density regions. Moreover, we introduce a geometry-aware alignment loss that minimizes geodesic discrepancy between source and target manifolds. Experiments on DomainNet, VisDA, and Office-Home show that MAADA consistently outperforms existing adversarial and adaptation methods in both unsupervised and few-shot settings, demonstrating superior structural robustness and cross-domain generalization. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_15191 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Geometrically Regularized Transfer Learning with On-Manifold and Off-Manifold Perturbation Satou, Hana Mitkiy, Alan Collins, Emma Kingston, Finn Computer Vision and Pattern Recognition Transfer learning under domain shift remains a fundamental challenge due to the divergence between source and target data manifolds. In this paper, we propose MAADA (Manifold-Aware Adversarial Data Augmentation), a novel framework that decomposes adversarial perturbations into on-manifold and off-manifold components to simultaneously capture semantic variation and model brittleness. We theoretically demonstrate that enforcing on-manifold consistency reduces hypothesis complexity and improves generalization, while off-manifold regularization smooths decision boundaries in low-density regions. Moreover, we introduce a geometry-aware alignment loss that minimizes geodesic discrepancy between source and target manifolds. Experiments on DomainNet, VisDA, and Office-Home show that MAADA consistently outperforms existing adversarial and adaptation methods in both unsupervised and few-shot settings, demonstrating superior structural robustness and cross-domain generalization. |
| title | Geometrically Regularized Transfer Learning with On-Manifold and Off-Manifold Perturbation |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2505.15191 |