Speaker
Description
We discuss an AI-assisted inference pipeline that learns the mapping between Wilson coefficient deformations and collider observables directly from simulated event data. This approach constructs physics-informed latent representations of SMEFT parameter space using neural architectures trained on differential cross sections and kinematic distributions. These learned embeddings enable efficient anomaly detection, parameter discrimination, and dimensionality reduction while preserving sensitivity to interference effects and operator correlations. We demonstrate how the framework accelerates likelihood-free inference, improves separation between Standard Model and BSM hypotheses, and provides interpretable structure in the space of EFT deformations.
| Presentation type. | Talk |
|---|