New release

[changhongyan123]

🚀 New Features and Enhancements:

• Support reference attacks that use the IN models and OUT models (e.g., Lira attack from Carlini et al., 2022.)
• Provide the benchmark results.
• Add validation results for evaluating the correctness of privacy meter implementation.
• Integrated hlb-CIFAR10 library into Privacy Meter. This integration enables running the attack with 16 shadow models and 1 target model in 10 minutes using a single NVIDIA TITAN RTX GPU.
• Added support for CIFAR100, Texas dataset, and Purchase Dataset.
• Added support for commonly used models, e.g., ResNet, VGG, etc.
• Enable customized augmented membership queries.
• Provide examples for different games in Ye et al., 2022.
• Support different attack signals, e.g., loss and rescaled logits.
• Provide a video tutorial.