This directory contains example scripts to help you get started with CertifiedAttack.

An interactive script that guides you through running different attack scenarios:

python examples/quick_start.py --demo

This will present you with a menu of pre-configured attack examples to choose from.

A minimal example showing how to run attacks programmatically:

# Basic usage
python examples/simple_attack.py

# Specify model and dataset
python examples/simple_attack.py --model vgg --dataset cifar100

# Different attack
python examples/simple_attack.py --attack PointWise

Compare multiple attacks on the same model:

python examples/compare_attacks.py --model resnet --dataset cifar10

Evaluate attacks against different defense mechanisms:

python examples/evaluate_defenses.py --defense blacklight

Example of creating and using custom configurations:

python examples/custom_config_example.py

For interactive exploration, check out our Jupyter notebooks:

• notebooks/01_getting_started.ipynb - Introduction to CertifiedAttack
• notebooks/02_attack_visualization.ipynb - Visualizing attack results
• notebooks/03_defense_comparison.ipynb - Comparing defense mechanisms
• notebooks/04_custom_attacks.ipynb - Implementing custom attacks

1. Ensure you have trained models in the experiments/ directory:

   python train.py --config configs/cifar10/resnet.yaml

2. Install all dependencies:

   pip install -r requirements.txt

3. Run from the project root directory:

   cd /path/to/CertifiedAttack
   python examples/simple_attack.py

• No checkpoint found: Train a model first using the training scripts
• CUDA out of memory: Reduce batch size in the examples
• Import errors: Make sure you're running from the project root directory

To create custom examples:

1. Copy one of the existing examples as a template
2. Modify the configuration parameters
3. Add your custom logic
4. Save in the examples directory

Example structure:

import sys
sys.path.append('..')  # Add parent directory to path

from pytorch_image_classification import create_model, create_dataloader
from attacks import get_attack

# Your custom code here

• Check the Usage Guide for detailed documentation
• Review the configuration files for more options
• Open an issue on GitHub for specific questions