Skip to content. | Skip to navigation

Sections
Personal tools
You are here: Home Theses Investigating ML-Ops tools for developing data-driven software components.

Contact

Prof. Dr. S. Decker
RWTH Aachen
Informatik 5
Ahornstr. 55
D-52056 Aachen
Tel +49/241/8021501
Fax +49/241/8022321

How to find us

Annual Reports

Disclaimer

Webmaster

 

 

Investigating ML-Ops tools for developing data-driven software components.

Thesis type
  • Master
Status Running
Submitted in 2022
Supervisor(s)
Advisor(s)

With the advent of digitization, more and more data-driven software components are being developed and brought into production. The functionality of data-driven software components is not entirely defined by the programmer in the classical way (i.e., by programming it directly), but relies upon data (e.g., through learning). This raises new challenges at the intersection between software engineering and data science. For example, the development life cycle of a data-driven component involves specific "data science" steps, (e.g. data preparation, feature engineering or model selection, see Figure 1). These steps produce artifacts (e.g., datasets, hyperparameters, models, or performance metrics) that must be properly managed.

Currently existing DevOps methods and tools have been recently adapted to meet the need of Data Science and gave rise to terms like ML-Ops, Data-Ops, or AI-Ops (Karamitsos et al. 2020; Mäkinen et al. 2021) and several new tools have entered the market (like, DVC, ML-Flow, etc.). However, how they differ from each other, what their main strengths and weaknesses are ,and which one should be chosen depending on the purpose and context of the application has not yet been systematically investigated. [tentative]

Document Actions