JACoW is a publisher in Geneva, Switzerland that publishes the proceedings of accelerator conferences held around the world by an international collaboration of editors.
@inproceedings{leputa:icalepcs2023-tu1bco01,
author = {M.F. Leputa and K.R.L. Baker and M. Romanovschi},
title = {{A Workflow for Training and Deploying Machine Learning Models to EPICS}},
% booktitle = {Proc. ICALEPCS'23},
booktitle = {Proc. 19th Int. Conf. Accel. Large Exp. Phys. Control Syst. (ICALEPCS'23)},
eventdate = {2023-10-09/2023-10-13},
pages = {244--248},
paper = {TU1BCO01},
language = {english},
keywords = {controls, EPICS, GPU, framework, software},
venue = {Cape Town, South Africa},
series = {International Conference on Accelerator and Large Experimental Physics Control Systems},
number = {19},
publisher = {JACoW Publishing, Geneva, Switzerland},
month = {02},
year = {2024},
issn = {2226-0358},
isbn = {978-3-95450-238-7},
doi = {10.18429/JACoW-ICALEPCS2023-TU1BCO01},
url = {https://jacow.org/icalepcs2023/papers/tu1bco01.pdf},
abstract = {{The transition to EPICS as the control system for the ISIS Neutron and Muon Source accelerators is an opportunity to more easily integrate machine learning into operations. But developing high quality machine learning (ML) models is insufficient. Integration into critical operations requires good development practices to ensure stability and reliability during deployment and to allow robust and easy maintenance. For these reasons we implemented a workflow for training and deploying models that utilize off-the-shelf, industry-standard tools such as MLflow. Our experience of how adoption of these tools can make developer’s lives easier during the training phase of a project is discussed. We describe how these tools may be used in an automated deployment pipeline to allow the ML model to interact with our EPICS ecosystem through Python-based IOCs within a containerized environment. This reduces the developer effort required to produce GUIs to interact with the models within the ISIS Main Control Room as tools familiar to operators, such as Phoebus, may be used. }},
}