bound_propagation
Linear and interval bound propagation in PyTorch
bound_propagation is a PyTorch library for linear and interval bound propagation with an extensible module-based API.
Features
- Support for IBP, CROWN, and CROWN-IBP bound propagation
- Alpha-CROWN support via projected gradient descent optimization
- GPU-enabled workflows and batched tensor operations in PyTorch
- Extensible BoundModule factory design for custom neural modules
Applications
- Certified robustness analysis
- Building custom bound-propagation pipelines for new model components