MLManagement: automating ML pipelines with reusable abstractions and access control

MLManagement: automating ML pipelines with reusable abstractions and access control

Ryndin M.A. (ISP RAS, Moscow, Russia)
Boiko A.M. (ISP RAS, Moscow, Russia)
Kushchuk D.O. (ISP RAS, Moscow, Russia)
Kiranov D.M. (ISP RAS, Moscow, Russia)

Abstract

The current Machine Learning (ML) landscape requires engineers to train and deliver models robustly, timely and at scale to stay competitive. Time to market could be reduced if specialists had a tool that would automate and address ML engineering challenges specific to the ML workflow such as experiment tracking, model versioning, etc. Although existing open-source MLOps tools provide functionality that partially covers these needs, most of them have limited access control and experiment code reuse capabilities. To this end we introduce the MLManagement platform. Our main contributions are as follows: firstly, we introduce reusable building block abstractions that enable users to quickly create automated ML experiments in a “plug-and-play” fashion. Secondly, we encourage collaboration in big teams or multi-team organizations by implementing Attribute-Based Access Control (ABAC) for all the platform's entities such as models, artifacts and data, as well as hardware resources allocation.

Keywords

MLOPS; machine learning; access control.

Edition

Proceedings of the Institute for System Programming, vol. 38, issue 3, part 1, 2026, pp. 143-152

ISSN 2220-6426 (Online), ISSN 2079-8156 (Print).

DOI: 10.15514/ISPRAS-2026-38(3)-8

For citation

Ryndin M.A., Boiko A.M., Kushchuk D.O., Kiranov D.M. MLManagement: automating ML pipelines with reusable abstractions and access control. Proceedings of the Institute for System Programming, vol. 38, issue 3, part 1, 2026, pp. 143-152 DOI: 10.15514/ISPRAS-2026-38(3)-8.

Full text of the paper in pdf Back to the contents of the volume