Abstract

The following research aims to examine the durability of data-use purposes through AI data pipelines. The research investigates if constraints related to the source data may carry through to derivation, generation, model building, training, and output processing while retaining their association with the regulated source. An internally developed data set of 117 cases was created and each case included a stated purpose, a potential use condition, a restriction use marker, provenance information, and a use that was expected to occur downstream. Support for the pipeline simulation, tracking of records, validation and production of graphs were achieved using Python, and document and spreadsheet tooling was used to facilitate structured reporting and consistency checks. The envisioned component architecture stores purpose metadata next to each data object, and validates it prior to taking it to the next processing step. The results show that explicit constraint packaging is better at maintaining intended use when compared to informal handling of stages, in particular for restricted use records and model facing representations. The study also shows that embeddings and derived features should retain a traceable relationship to the source policy while being governed as transformed representations rather than as identical copies of the original content. The contribution is a practical persistence pattern, accepting that purpose is actually an operational metadata, not an ad hoc collection notice and establishes a traceable path from data ingestion to model output.

References

• T. Akidau et al., “The dataflow model: A practical approach to balancing correctness, latency, and cost in massive-scale, unbounded, out-of-order data processing,” Proc. VLDB Endowment, vol. 8, no. 12, pp. 1792–1803, 2015. https://doi.org/10.14778/2824032.2824076
• C. Batini and M. Scannapieco, Data and Information Quality: Dimensions, Principles and Techniques. Cham, Switzerland: Springer, 2016. https://doi.org/10.1007/978-3-319-24106-7
• D. Baylor et al., “TFX: A TensorFlow-based production-scale machine learning platform,” in Proc. 23rd ACM SIGKDD Int. Conf. Knowledge Discovery and Data Mining (KDD), Halifax, NS, Canada, 2017, pp. 1387–1395. https://doi.org/10.1145/3097983.3098021
• E. Breck et al., “Data validation for machine learning,” in Proc. 2nd Conf. Machine Learning and Systems (SysML), Stanford, CA, USA, 2019. [Online]. Available: https://research.google/pubs/data-validation-for-machine-learning/
• E. Caveness et al., “TensorFlow Data Validation: Data analysis and validation in continuous ML pipelines,” in Proc. 2020 ACM SIGMOD Int. Conf. Management of Data, Portland, OR, USA, 2020, pp. 2793–2796. https://doi.org/10.1145/3318464.3384707
• M. Koehler et al., “Incorporating data context to cost-effectively automate end-to-end data wrangling,” IEEE Trans. Big Data, vol. 7, no. 1, pp. 169–186, 2021. https://doi.org/10.1109/TBDATA.2019.2907588
• A. Debrouvier et al., “A model and query language for temporal graph databases,” VLDB J., vol. 30, no. 5, pp. 825–858, 2021. https://doi.org/10.1007/S00778-021-00675-4
• I. F. Ilyas and T. Rekatsinas, “Machine learning and data cleaning: Which serves the other?” ACM J. Data Inf. Quality, vol. 14, no. 3, 2022. https://doi.org/10.1145/3506712
• Y. Gong et al., “A survey on dataset quality in machine learning,” Inf. Softw. Technol., vol. 162, 2023. https://doi.org/10.1016/j.infsof.2023.107268
• DAMA International, “DAMA-DMBOK: Data Management Body of Knowledge,” 2024. [Online]. Available: https://www.dama.org/cpages/body-of-knowledge
• M. Garouani, F. Ravat, and N. Vallès-Parlangeau, “Model lake: A new alternative for machine learning models management and governance,” in Proc. Int. Conf. Web Information Systems Engineering (WISE), vol. 15439, Lecture Notes in Computer Science, pp. 133–144. Springer, 2024. https://doi.org/10.1007/978-981-96-0573-6_10
• R. Boselli, S. D’Amico, and N. Nobani, “eXplainable AI for word embeddings: A survey,” Cogn. Comput., vol. 17, no. 1, Art. no. 19, 2025. https://doi.org/10.1007/S12559-024-10373-2
• S. Sultan and C. D. Jensen, “Ensuring purpose limitation in large-scale infrastructures with provenance-enabled access control,” Sensors, vol. 21, no. 9, Art. no. 3041, 2021. https://doi.org/10.3390/s21093041
• J. Park and R. Sandhu, “The UCONABC usage control model,” ACM Trans. Inf. Syst. Security, vol. 7, no. 1, pp. 128–174, 2004. https://doi.org/10.1145/984334.984339
• R. Iannella and S. Villata, “ODRL Information Model 2.2,” W3C Recommendation, Feb. 15, 2018. https://www.w3.org/TR/odrl-model/
• T. Lebo, S. Sahoo, and D. McGuinness, Eds., “PROV-O: The PROV Ontology,” W3C Recommendation, Apr. 30, 2013. https://www.w3.org/TR/prov-o/
• C. Jung and J. Dörr, “Data usage control,” in Designing Data Spaces, Springer, 2022, pp. 129–146. https://doi.org/10.1007/978-3-030-93975-5_8