Deep Networks for HEP

This page provides example code, datasets and recipes for running HEP Physics analyses using deep neural networks on Cori. The current scripts were those used for the CNN classification and timing studies reported at this ACAT paper.

Datasets

These contain simulated data with an ATLAS-like detector. Data is available from http://portal.nersc.gov/project/mpccc/wbhimji/RPVSusyData/ . A README is provided in the directory.

Convolutional Neural Network for Classification

This provides a network for classification (RPVSusy signal vs QCD background) on 3-channel (calorimeter + track) whole-detector images as described in the ACAT paper. It implements 3 convolution+pooling units with rectified linear unit (ReLU) activation functions. These layers output into two fully connected layers, with cross-entropy as the loss function and the ADAM optimizer.

Code

The Keras code to implement the CNN used in the paper is available at https://github.com/eracah/atlas_dl/tree/micky . This single script is fairly self explanatory and easily run at NERSC.

Code for preselection of data as well as for Lasgne/Theano implementations is in the main branch of that repository.

Code for shallow classifiers compared in the paper is available at https://github.com/sparticlesteve/acat2017-rpvdl-shallow.

Note

A more recent tensorflow implementation is used in the NERSC science benchmarks