Projects

Projects that have repositories with code:

Numerical Integration with Normalizing Flows: i-flow

We use normalizing flows to improve importance sampling in numerical integration. Here, we have our code i-flow that implements this idea and here we describe it in detail.

Simulating Calorimeter Showers with Normalizing Flows: CaloFlow

We use normalizing flows as deep generative model to simulate calorimeter showers of positrons, photons, and pions. Our code CaloFlow is available on gitlab. CaloFlow v1, which implements sampling via a MAF is described here; CaloFlow v2, which implements sampling via an IAF and is trained with probability density distillation is described here. This project is the first application of normalizing flows to detector simulation and the first deep generative model that passes the classifier test of distinguishing real (based on GEANT4) from fake (based on the surrogate).

Enhancing anomaly detection in bump hunts: CATHODE

In Classifying Anomalies THrough Outer Density Estimation (CATHODE), we use normalizing flows to enhance anomaly searches based on bump hunts. Using the LHC Olympics 2020 R&D dataset, we show that CATHODE can enhance potential signals up to the theoretical maxium, set by a classifier trained to distinguish perfectly modeled background and data. The publice repository of this project is here, the paper describing the method is available here.

Study of Electroweak Symmetry Non-Restoration

In this paper we compute the effective potential of an extended Two-Higgs-Doublet model and study electroweak symmetry non-restoration at high temperatures. The code for the numerical evaluation can be found here.