Robustness Verification for Transformers

This project investigates sound and scalable approaches to formal verification for Transformer models. The work focuses on reasoning about attention operations, constructing useful bounds for nonlinear components, and certifying robustness properties.