Machine Learning Pipelines with Modern Big Data Tools for High Energy Physics
This work addresses the problem of scaling ML workflows for high energy physics researchers, but it is incremental as it combines existing tools rather than introducing new methods.
The paper tackles the challenge of implementing end-to-end machine learning pipelines for high energy physics by integrating big data tools like Apache Spark with HEP-specific data formats and distributed training frameworks, achieving scalable neural network training on Spark clusters.
The effective utilization at scale of complex machine learning (ML) techniques for HEP use cases poses several technological challenges, most importantly on the actual implementation of dedicated end-to-end data pipelines. A solution to these challenges is presented, which allows training neural network classifiers using solutions from the Big Data and data science ecosystems, integrated with tools, software, and platforms common in the HEP environment. In particular, Apache Spark is exploited for data preparation and feature engineering, running the corresponding (Python) code interactively on Jupyter notebooks. Key integrations and libraries that make Spark capable of ingesting data stored using ROOT format and accessed via the XRootD protocol, are described and discussed. Training of the neural network models, defined using the Keras API, is performed in a distributed fashion on Spark clusters by using BigDL with Analytics Zoo and also by using TensorFlow, notably for distributed training on CPU and GPU resourcess. The implementation and the results of the distributed training are described in detail in this work.