Progress on cloud native solution of Machine Learning as a Service for HEP - Update on INFN Cloud prototype

Aug 4, 2026

Upload by Violet

Machine Learning (ML) has become valuable in High-Energy Physics (HEP) and will face escalating demands with the High-Luminosity LHC upgrade, where exascale-scale data volumes will stress storage, processing, and analysis. A practical barrier is the effort required to build ML pipelines that can read and train using ROOT data formats while supporting remote and distributed sources. The MLaaS4HEP approach provides an experiment-agnostic service for end-to-end ML pipelines—data access, processing, training, and inference—implemented in a cloud-native architecture. Updates are presented, including a working INFN Cloud prototype integrating OAuth2 authentication, XRootD access, and TFaaS inference.

LanguageEnglish

CategoryTechnology

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Introduction

Motivation for ML in HEP

MLaaS4HEP and system overview

The developed MLaaS solution for HEP

The MLaaS4HEP framework and training workflow

Progress on cloud native solution of Machine Learning asa Service for HEP Luca Giommi1 , ∗ , Daniele Spiga2 , Valentin Kuznetsov3 , and Daniele Bonacorsi4,5 1INFN-CNAF, Viale Carlo Berti Pichat, 6/2, 40127 Bologna (ITALY) 2INFN Sezione di Perugia, Via Alessandro Pascoli 23c, 06123 Perugia (ITALY) 3Cornell University, 616 Thurston Ave., Ithaca, NY 14853 (USA) 4University of Bologna, Via Zamboni 33, 40126 Bologna (ITALY) 5INFN Sezione di Bologna, Viale Carlo Berti Pichat, 6/2, 40127 Bologna (ITALY) Abstract. Nowadays Machine Learning (ML) techniques are successfully used in many areas of High-Energy Physics (HEP) and will play a significant role also in the upcoming High-Luminosity LHC upgrade foreseen at CERN, when a huge amount of data will be produced by LHC and collected by the experiments, facing challenges at the exascale. To favor the usage of ML in HEP analyses, it would be useful to have a service allowing to perform the entire ML pipeline (in terms of reading the data, processing data, training a ML model, and serving predictions) directly using ROOT files of arbitrary size from local or remote distributed data sources. The Machine Learning as a Service for HEP (MLaaS4HEP) solution we have already proposed aims to provide such kind of service and to be HEP experiment agnostic. To provide users with a real service and to integrate it into the INFN Cloud, we started working on MLaaS4HEP cloudification. This would allow to use cloud resources and to work in a distributed environment. In this work, we provide updates on this topic and discuss a working prototype of the service running on INFN Cloud. It includes an OAuth2 proxy server as authentication/authorization layer, a MLaaS4HEP server, an XRootD proxy server for enabling access to remote ROOT data, and the TensorFlow as a Service (TFaaS) service in charge of the inference phase. With this architecture a HEP user can submit ML pipelines, after being authenticated and authorized, using local or remote ROOT files simply using HTTP calls. 1 Introduction The combined operations of the Large Hadron Collider (LHC) experiments yield approximately 200 PB of data annually, necessitating to be stored, processed, and analyzed. To enable physicists spread all over the world to access the required computing power and storage, CERN employs a grid-based network called Worldwide LHC Computing Grid (WLCG) . The upcoming High Luminosity LHC (HL-LHC) program, scheduled to start in 2029, will amass around 1 EB of data yearly from ATLAS and CMS, to which derived and simulated ∗ e-mail: [luca.giommi@cnaf.infn.it](luca.giommi@cnaf.infn.it) © The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0 ([https://creativecommons.org/licenses/by/4.0/](https://creativecommons.org/licenses/by/4.0/)). data is added. This presents a significant challenge, as each experiment will enter a multiExabyte per year data management regime. In this scenario, the role of Machine Learning (ML) in High Energy Physics (HEP) will be critical. ML techniques have found success across various areas of HEP, including online and offline reconstruction, detector simulation, object identification, and Monte Carlo generation, among others. However, developing and implementing ML projects for practical use is timeintensive, demanding specific skills, and HEP analysts often lack the necessary data science expertise to tackle these challenges independently. Compounding this issue is the gap between the HEP and ML communities. This is partially due to the prevalent use of the ROOT [1] data format within HEP, which remains largely unfamiliar outside this domain. HEP data relies on tree-based structures, where the event size is unpredictable (e.g. the particle count may vary in each physics event) so that careful consideration is needed when using ROOT data with ML frameworks. Therefore, offering a service to H

FAQs

What problem does MLaaS4HEP aim to solve for HEP users?

MLaaS4HEP provides a service that runs the full ML pipeline—reading ROOT data, processing, training, and serving predictions—so users can execute ML workflows with local or remote data sources more easily.

How does the architecture enable access to remote ROOT files?

The prototype uses an XRootD proxy server to enable access to remote ROOT datasets, allowing pipelines to operate on distributed data sources.

What components are included in the INFN Cloud prototype?

The service includes an OAuth2 proxy for authentication/authorization, an MLaaS4HEP server for pipeline execution, an XRootD proxy for remote ROOT access, and TFaaS for inference through TensorFlow-hosted models.

Aug 4, 2026

Upload by Violet