ELSA - Enhanced Latent SpAces for improved collider simulations

Introduction

This repo contains the code for the Enhanced Latent SpAces (ELSA) framework for neural network improved collider simulations. It is based on the LaSeR protocol [1] and further employs augmented flows [2]. The code is provided in PyTorch.

Deprecation warning

The mappings (Rambo etc) in this reporsitory are deprecated and have been superseded by the MadSpace framework: Link to repo

Installation

Dependencies

Package	Version
Python	>= 3.7
Torch	>= 1.8
Numpy	>= 1.20.0

Download + Install

# clone the repository
git clone https://github.com/ramonpeter/elsa.git
# then install in dev mode
cd elsa
python setup.py develop

Prepare datasets

In order to download the datasets

./get_datasets.sh

This prepares and/or downloads the datasets into the datasets folder.

References

[1] Latent Space Refinement: 2106.00792

• Code: https://github.com/ramonpeter/LaSeR

[2] SurVAE Flows: 2007.02731

• Code https://github.com/didriknielsen/survae_flows