Machine Learning with ROOT/TMVA

Kim Albertsson,Sitong An,Lorenzo Moneta,Omar Andres Zapata Mesa,Joana Niermann,Sergei Gleyzer,Stefan Wunsch,Luca Zampieri,C Doglioni,G.A Stewart,D Kim,W Kamleh,P Jackson,L Silvestris

doi:10.1051/epjconf/202024506019

Abstract

ROOT provides, through TMVA, machine learning tools for data analysis at HEP experiments and beyond. We present recently included features in TMVA and the strategy for future developments in the diversified machine learning landscape. Focus is put on fast machine learning inference, which enables analysts to deploy their machine learning models rapidly on large scale datasets. The new developments are paired with newly designed C++ and Python interfaces supporting modern C++ paradigms and full interoperability in the Python ecosystem. We present as well a new deep learning implementation for convolutional neural network using the cuDNN library for GPU. We show benchmarking results in term of training time and inference time, when comparing with other machine learning libraries such as Keras/Tensorflow.

Highlights

The ROOT data analysis toolkit [1] provides, through TMVA [2], a large amount of machinelearning methods for data analysis in HEP and beyond, which are collected in the package since 2005
This work presents the new developments in ROOT/TMVA and discusses how the project positions itself in the machine-learning landscape of today
This changes the focus of ROOT/TMVA for modern neural network architectures from providing the algorithms itself to being the glue between the third-party machine learning libraries and the software environment in HEP experiments

Summary

Introduction

The ROOT data analysis toolkit [1] provides, through TMVA [2], a large amount of machinelearning methods for data analysis in HEP (high-energy particle physics) and beyond, which are collected in the package since 2005. Did the academic field change, due to the applicability of these methods in industry, the software landscape evolved quickly and is today dominated by products maintained by large technology companies [8,9,10]. These modern machine-learning methods and software tools have been adopted by the HEP community since early on [11] and is today successfully in production, for example in the CMS DeepJet tagger [12] or the ATLAS quark-gluon tagger [13]. This work presents the new developments in ROOT/TMVA and discusses how the project positions itself in the machine-learning landscape of today

Interoperability with the machine learning ecosystem

Modernization of TMVA

Fast decision tree inference

Fast neural networks

Outlook