MLaaS4HEP: Machine Learning as a Service for HEP
Valentin Kuznetsov, Luca Giommi, Daniele Bonacorsi
Machine Learning (ML) will play a significant role in the success of the upcoming High-Luminosity LHC (HL-LHC) program at CERN. An unprecedented amount of data at the Exa-Byte scale will be collected by LHC experiments in the next decade, and this effort will require novel approaches to train and use ML models. In this paper, we discuss a Machine Learning as a Service pipeline (MLaaS4HEP) which provides three independent layers: a data streaming layer to read High-Energy Physics (HEP) data in their native ROOT data format; a data training layer to train ML models using distributed ROOT files; a data inference layer to serve pre-trained model via HTTP protocol. Such modular design opens up the possibility to train data at large scale by reading ROOT files from remote storages, e.g. World-Wide LHC Grid (WLCG) infrastructure, and avoid data-transformation step to flatten the data to data-formats currently used by ML frameworks. In addition, it may provide an easy access to pre-trained ML models in existing infrastructure and applications. In particular, we demonstrate the usage of the MLaaS4HEP architecture for a concrete physics use-case based on tt¯ Higgs analysis in CMS. We provide details on how we train the ML model using distributed ROOT files, and discuss the performance of the MLaaS approach for the selected physics analysis.