Capturing end-to-end provenance for machine learning pipelines

IF 3 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Systems Pub Date : 2024-12-22 DOI:10.1016/j.is.2024.102495

Marius Schlegel, Kai-Uwe Sattler

引用次数: 0

Abstract

Modern workflows for developing ML pipelines utilize ML artifact management systems (ML AMSs) such as MLflow in addition to traditional version control systems such as Git. ML AMSs collect data, model, metadata and software artifacts used and produced in pipeline development workflows. While ensuring repeatability and reproducibility, the provenance capabilities are still rudimentary, mainly due to incomplete traces, coarse granularity, and limited query capabilities. In this paper, we introduce a comprehensive PROV-compliant provenance model that captures end-to-end provenance traces of ML pipelines, their artifacts, and their relationships based on MLflow and Git activities. Moreover, we present the tool MLflow2PROV for continuously extracting provenance graphs according to our model, enabling querying, analyzing, and processing of the collected provenance information.

查看原文本刊更多论文

求助全文

约1分钟内获得全文求助全文

来源期刊

Information Systems 工程技术-计算机：信息系统

CiteScore

9.40

自引率

2.70%

发文量

112

审稿时长

53 days

期刊介绍： Information systems are the software and hardware systems that support data-intensive applications. The journal Information Systems publishes articles concerning the design and implementation of languages, data models, process models, algorithms, software and hardware for information systems. Subject areas include data management issues as presented in the principal international database conferences (e.g., ACM SIGMOD/PODS, VLDB, ICDE and ICDT/EDBT) as well as data-related issues from the fields of data mining/machine learning, information retrieval coordinated with structured data, internet and cloud data management, business process management, web semantics, visual and audio information systems, scientific computing, and data science. Implementation papers having to do with massively parallel data management, fault tolerance in practice, and special purpose hardware for data-intensive systems are also welcome. Manuscripts from application domains, such as urban informatics, social and natural science, and Internet of Things, are also welcome. All papers should highlight innovative solutions to data management problems such as new data models, performance enhancements, and show how those innovations contribute to the goals of the application.