Study of ML-Enabled Software Systems

ML-enabled Software Systems

Due to the rapid advances in AI, the question of suitable applications for new models always arises. Specifically, machine learning (ML)-enabled systems are of interest, as approximately 80% of such projects fail or stall in production.

Our research team aims to investigate this issue and determine how to improve the adoption of ML into software systems by learning from current projects and research.

Mining Study

To gain insight into the current state of machine learning (ML)-enabled systems, we conducted a mining study of 2,928 open-source systems. Our analysis revealed that, although many systems adopt ML, they often lack a structured approach. Specifically, we found that ML model reuse often involves copying existing implementations.

Additionally, we analyzed the source code of 26 systems that use ML to implement end–user–oriented functionalities. Our analysis revealed that these systems use multiple ML models, which can be structured into five interaction patterns.

For more details on the current state of the integration of ML models into software systems and ML-asset reuse practices, take a look at our paper here.

Survey

Based on the findings of our mining study, we are now conducting a survey to gather the perspectives of practitioners on this topic.
We invite practitioners involved with ML-enabled systems, such as developers, data scientists, or project managers, to take part in our survey on ML-enabled software systems. Your benefit is to reflect about your practices, learn about others’ practices, and later receive a state-of-practice report with the results of this survey. 
The survey takes only 15 minutes to fill in. It contains questions on the ML technologies used in systems, on the integration of ML models into software and the reuse of ML models. 
 
You can access the survey here: TBA
 
 

Publications

  • Sens, Yorick; Knopp, Henriette; Peldszus, Sven; Berger, Thorsten: A Large-Scale Study of Model Integration in ML-Enabled Software Systems. In: Proceedings of the 47th International Conference on Software Engineering (ICSE), 2025, https://doi.org/10.1109/ICSE55347.2025.00185