STN-Code

KO | EN

gitlite — search

#hacktoberfest #react #ai #llm #machine-learning #rust #deep-learning #docker #java #nodejs #windows #awesome #vue #security

(★ 22)

Code for Stability Training with Noise (STN)

STN-Code 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Stability Training with Noise (STN)

This repository contains code to reproduce results from the paper:

**Certified Adversarial Robustness with Additive Gaussian Noise** <br>
*Bai Li, Changyou Chen, Wenlin Wang, Lawrence Carin* <br>
ArXiv report: https://arxiv.org/abs/1809.03113

###### REQUIREMENTS

The code was tested with Python 3.6.8, Tensorflow 1.8.0 and Keras 2.2.4.

To compute certified bound, install R, rpy2 and R package *MultinomialCI*.

###### EXPERIMENTS

To evaluate robustness, we use

```
python -m simple_eval [dataset] [attack] [source_model] [target_model] [norm] [size]
```

For Example, to run attacks on CIFAR10 with l2 attacks bounded by 4

```
python -m simple_eval cifar10 so models/model_CIFAR10 models/modelA_CIFAR10 l2 4

```

It reports the classification accuracy of model_CIFAR10 against white-box attack and transferred attack from modelA_CIFAR10.

When l2 norm is selected, we use the Carlini and Wagner attack. When linf norm is selected, we use the PGD attack.

To compute the certified robust accuracy, we use 
```
python -m bound [dataset] [source_model] [size]
```

For example
```
python -m bound mnist models/model_mnist 1.2
```

It reports the lower bound of the classification accuracy of model_mnist against arbitrary attacks with l2 norm no greater than 1.2.

STN-Code

STN-Code 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

STN-Code

STN-Code 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README