# CURE_robustness

**Repository Path**: purple_ai/CURE_robustness

## Basic Information

- **Project Name**: CURE_robustness
- **Description**: 基于曲率优化提升模型鲁棒性
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2025-03-13
- **Last Updated**: 2025-03-13

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# CURE for adversarial vulnerability
CURE is a deep network training algorithm via a curvature regularizer. Networks trained using CURE can achieve significant adversarial robustness.

### Dependencies
---
The code is compatible with python3.7. To install the dependecies use
```
pip install -e .
```

To start, you can play with the following notebook:

* "[CURE Example Code](https://github.com/F-Salehi/CURE_robustness/blob/master/notebooks/example.ipynb) "


### Reference 
----
"[Robustness via curvature regularization, and vice versa](https://arxiv.org/abs/1811.09716) ", SM. Moosavi-Dezfooli, A. Fawzi, J. Uesato, and P. Frossard, _CVPR 2019_.