DAG-oriented representation of model synthesis algorithm construction processes by monitoring software agents
DOI:
https://doi.org/10.34121/1028-9763-2026-1-73-86Keywords:
monitoring, software agent, models, directed acyclic graph (DAG), reproducibility, decomposition, model synthesis algorithms, model ensemblesAbstract
The article considers a fundamental task of ensuring reliable and reproducible construction of multilayer model synthesis algorithms (MSA) intended for use in monitoring software agents as part of multi-agent systems (MAS). After successful training of multilayer model ensembles and other complex MSAs that demonstrate high accuracy in a research environment, a challenge arises related to their portable and deterministic deployment. To solve this problem, a DAG-oriented architecture is proposed, where the entire logic of the MSA execution is represented as a single, self-sufficient, and easily interpreted directed acyclic graph (DAG). A set of typed graph nodes has been developed that allows for transparent representation of complex operations such as increasing data homogeneity and multilayer recirculation, breaking them into logical and understandable steps. The process of transforming the trained architecture into a portable DAG containing all dependencies between components, paths to artifacts, and aggregation parameters is investigated. It is demonstrated that the proposed approach, due to clearly defined rules of data exchange between nodes and a fixed execution order guaranteed by topological sorting, ensures full compliance of the execution results with those obtained during training. It is established that such an architecture creates a reliable bridge between the research environment and practical application. It provides portability, transparency (thanks to tracing tools), and verifiability (due to the possibility of evaluating intermediate nodes) of complex methods of MSA construction. In conclusion, the developed approach is important for building reliable software agents capable of stable and predictable functioning in real-world conditions.
References
1. Model persistence. scikit-learn. URL: https://scikit-learn/stable/model_persistence.html (дата звернення: 19.10.2025).
2. Guazzelli A., Zeller M., Lin W.C. та ін. PMML: An open standard for sharing models. R Journal. Вип. 1, N 1. С. 60–65. DOI: https://doi.org/10.32614/rj-2009-010.
3. Wooldridge M. An Introduction to MultiAgent Systems. John Wiley & Sons, 2009. 484 с. ISBN 978-0- 470-51946-2.
4. Голуб С.В. Багаторівневе моделювання в технологіях моніторингу оточуючого середовища. Черкаси: Вид. від. ЧНУ імені Богдана Хмельницького, 2007. P. 220.
5. Breck E., Cai S., Nielsen E. et al. The ML test score: A rubric for ML production readiness and technical debt reduction. 2017 IEEE International Conference on Big Data (Big Data 2017 IEEE International Conference on Big Data (Big Data). P. 1123–1132. DOI: https://doi.org/10.1109/BigData.2017.8258038.
6. Dags — Airflow 3.1.0 Documentation. URL: https://airflow.apache.org/docs/apache-airflow/stable/coreconcepts/dags.html (accessed 19/10/2025).
7. Holub S.V., Ostapiuk V.V. Machine learning of multilayer models of a monitoring software agent. Mathematical machines and systems. 2025. Vol. 2. P. 76–96. DOI: https://doi.org/10.34121/1028-9763-2025-2-76-95.
8. ONNX Concepts — ONNX 1.20.0 documentation. URL: https://onnx.ai/onnx/intro/concepts.html (accessed 19/10/2025).
9. Rocklin M. Dask: Parallel Computation with Blocked algorithms and Task Scheduling. Python in Science Conference. Austin, Texas, 2015. P. 126–132. DOI: https://doi.org/10.25080/Majora-7b98e3ed-013.
10. Sculley D., Holt G., Golovin D. et al. Hidden Technical Debt in Machine Learning Systems. Advances in Neural Information Processing SystemsCurran Associates, Inc., 2015. URL: https://papers.nips.cc/paper_files/paper/2015/hash/86df7dcfd896fcaf2674f757a2463eba-Abstract.html (accessed: 25/10/2025).
11. Zaharia M., Xin R. S., Wendell P. et al. Apache Spark: a unified engine for big data processing. Commun. ACM. 2016, Vol. 59, Issue 11. P. 56–65. DOI: https://doi.org/10.1145/2934664.
12. Doshi-Velez F., Kim B. Towards A Rigorous Science of Interpretable Machine Learning. arXiv, 2017. DOI: https://doi.org/10.48550/arXiv.1702.08608.
13. Low Y., Gonzalez J., Kyrola A. et al. GraphLab: A New Framework For Parallel Machine Learning. arXiv, 2014. DOI: https://doi.org/10.48550/arXiv.1408.2041.
161
29
