Architectural Design Decisions for the Machine Learning Workflow

Architectural Design Decisions for the Machine Learning Workflow

Abstract

Bringing machine learning models to production is challenging as it is often fraught with uncertainty and confusion, partially due to the disparity between software engineering and machine learning practices, but also due to knowledge gaps on the level of the individual practitioner. We conducted a qualitative investigation into the architectural decisions faced by practitioners as documented in gray literature based on Straussian Grounded Theory and modeled current practices in machine learning. Our novel Architectural Design Decision model is based on current practitioner understanding of the topic and helps bridge the gap between science and practice, foster scientific understanding of the subject, and support practitioners via the integration and consolidation of the myriad decisions they face. We describe a subset of the Architectural Design Decisions that were modeled, discuss uses for the model, and outline areas in which further research may be pursued.

Grafik Top
Authors
  • Warnett, Stephen J.
  • Zdun, Uwe
Grafik Top
Projects
Grafik Top
Shortfacts
Category
Journal Paper
Divisions
Software Architecture
Subjects
Systemarchitektur Sonstiges
Software Engineering
Kuenstliche Intelligenz
Systemarchitektur Allgemeines
Journal or Publication Title
Computer 2022
Publisher
IEEE Computer Society
Page Range
pp. 40-51
Number
3
Volume
55
Date
March 2022
Export
Grafik Top