Demonstration of Geyser: Provenance Extraction and Applications over Data Science Scripts
Open accessAbstract
As enterprises have started developing and deploying complicated data science workloads at scale, the need for mechanisms that enable enterprise-grade data science (e.g., compliance or auditing) has become more pronounced. In this paper, we present Geyser, an extensible provenance system for data science workloads that can be used as a foundation for enterprise-grade data science. Our system supports both static and dynamic provenance, over a wide range of data science scripts, driven by a knowledge base of data science APIs. We demonstrate the wide applicability of the system using various industrial applications: provenance extraction, model compliance, model linting, model versioning, and poisoning detection. A video of the demonstration is available at https://aka.ms/geyserdemo.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
MLOps is viewed from systematic management of ML experiments and assets, leading to version control and experiment recording.
“MLOps is also viewed from the perspective of systematic management of ML experiments and related assets”
From A Multivocal Review of MLOps Practices, Challenges and Open Issues (2026) · cites this paper as [18]
The Introduction describes systematic provenance tracking for data science workflows, and the Model Versioning section uses provenance across runs to explain why models differ or remain the same. That supports the model-versioning aspect, but the paper does not frame MLOps generally as management of ML experiments and assets or describe experiment recording. (LLM, conf=0.82)
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
Adversarial Robustness Toolbox v1.0.0Preprint
Nicolae, M.-I., Sinn, M., Minh, T. N., Buesser, B., Rawat, A., Wistuba, M., Zantedeschi, V., Baracaldo, N., Chen, B., Ludwig, H., et al. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1807.01069 · Cited by 467 · see citing works
The W3C PROV family of specifications for modelling provenance metadata
Missier, P., Belhajjame, K., & Cheney, J. (2013) · International Conference on Extending Database Technology (EDBT) · doi:10.1145/2452376.2452478 · Cited by 202 · see citing works
Karvounarakis, G., Ives, Z. G., & Tannen, V. (2010) · ACM SIGMOD International Conference on Management of Data (SIGMOD) · doi:10.1145/1807167.1807269 · Cited by 192 · see citing works
Detecting Poisoning Attacks on Machine Learning in IoT Environments
Baracaldo, N., Chen, B., Ludwig, H., Safavi, A., & Zhang, R. (2018) · IEEE International Congress on Internet of Things (ICIOT) · doi:10.1109/iciot.2018.00015 · Cited by 81 · see citing works
Fine-grained lineage for safer notebook interactions
Macke, S., Gong, H., Lee, D. J., Head, A., Xin, D., & Parameswaran, A. (2021) · Proceedings of the VLDB Endowment · doi:10.14778/3447689.3447712 · Cited by 48 · see citing works
Smoke: Fine-grained Lineage at Interactive SpeedPreprint
Psallidas, F., & Wu, E. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1801.07237 · Cited by 47 · see citing works
Self-citationChapman, A. P., Lauro, L., Paolo Missier, B., & Torlone, R. (2022) · Proceedings of the VLDB Endowment · doi:10.14778/3554821.3554857 · Cited by 6 · see citing works
Flagged first, then by citations
Research works printed but not linked: 4
- Adriane Chapman, Luca Lauro, Paolo Missier, and Riccardo Torlone. 2022. DPDS: assisting data science with data provenance. Proceedings of the VLDB Endowment 15, 12 (2022), 3614-3617.
- Boris Glavic and Gustavo Alonso. 2009. Perm: Processing provenance and data on the same data model through query rewriting. In ICDE.
- Mohammad Hossein Namaki, Avrilia Floratou, Fotis Psallidas, Subru Krishnan, Ashvin Agrawal, Yinghui Wu, Yiwen Zhu, and Markus Weimer. 2020. Vamsa: Au- tomated Provenance Tracking in Data Science Scripts. In SIGKDD '20. 1542-1551.
- Fotis Psallidas, Yiwen Zhu, Bojan Karlas, Jordan Henkel, Matteo Interlandi, Subru Krishnan, Brian Kroth, Venkatesh Emani, Wentao Wu, Ce Zhang, Markus Weimer, Avrilia Floratou, Carlo Curino, and Konstantinos Karanasos. 2022. Data Science Through the Looking Glass: Analysis of Millions of GitHub Notebooks and ML.NET Pipelines. SIGMOD Record 51, 2 (jul 2022), 30-37.
Cited by
Sorted by citations
Capturing end-to-end provenance for machine learning pipelines
Schlegel, M., & Sattler, K.-U. (2024) · Information Systems · doi:10.1016/j.is.2024.102495 · Cited by 29 · see citing works
A Multivocal Review of MLOps Practices, Challenges and Open Issues
Eken, B., Pallewatta, S., Tran, N., Tosun, A., & Babar, M. A. (2026) · ACM Computing Surveys · doi:10.1145/3747346 · Cited by 33 · see citing works
APEX-DAG: Library and Language independent Pipeline EXtraction
Eggers, S., Żukowska, N., & Abedjan, Z. (2025) · Proceedings of the VLDB Endowment · doi:10.14778/3750601.3750675
- Authors
- Psallidas, F., Leszczynski, M. E., Namaki, M. H., Floratou, A., Agrawal, A., Karanasos, K., Krishnan, S., Subotic, P., Weimer, M., Wu, Y., & Zhu, Y.
- Affiliations
- Microsoft (United States) (US)
- Stanford University (US)
- Meta (United States) (US)
- Microsoft, Redmond, WA, USA
- Case Western Reserve University (US)
- ORCID
- Venue
- Companion of the 2023 International Conference on Management of Data
- Publisher
- ACM
- Published
- 2023-06-04
- Pages
- 123-126
- ISSN
- 0730-8078 (ISSN-L)
- OpenAlex
- W4379390776
- Cited by
- 3
- References
- 6
- Citations by year
- 2025: 2 · 2024: 1
- FWCI
- 2.11
- Type
- conference paper
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- ACM SIGMOD International Conference on Management of Data (SIGMOD) (publishedVersion)
- Topics
- Social Sciences › Decision Sciences › Information Systems and Management › Scientific Computing and Data Management
- Physical Sciences › Computer Science › Information Systems › Software Engineering Research
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Keywords
- model versioning, poisoning detection
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
