README.md

---
license: other
license_name: carso-adapted-rail-m
license_link: LICENSE
datasets:
- cifar10
- cifar100
- jeremyf/tiny-imagent-200
metrics:
- accuracy
pipeline_tag: image-classification
---

Pre-trained models for the paper *"Carefully Blending Adversarial Training and Purification Improves Adversarial Robustness"* (Ballarin et al., 2024)

**Developed by:** [Emanuele Ballarin](https://ballarin.cc/), [Alessio Ansuini](https://areasciencepark-rit.gitlab.io/lade/alessio.ansuini/), [Luca Bortolussi](https://ai-lab.units.it/?page_id=139)  
**Repository:** [`github.com/emaballarin/CARSO`](https://github.com/emaballarin/CARSO)  
**Paper:** [*"Carefully Blending Adversarial Training and Purification Improves Adversarial Robustness"* (Ballarin et al., 2024)](https://arxiv.org/abs/2306.06081)  

**Citation (BibTeX):**
```bibtex
@misc{ballarin2023carefully,
    title={Carefully Blending Adversarial Training and Purification Improves Adversarial Robustness},
    author={Emanuele Ballarin and Alessio Ansuini and Luca Bortolussi},
    year={2023},
    eprint={2306.06081},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}
```