EUVIMEDEUVIMED entdecken
Crossref

Feature Serving at Scale: Design, Migration, and Empirical Evaluation of a Production Feature Platform

Karthik Perikala

Journal of Artificial intelligence and Machine Learning · 2024 · 2 (1) · 1-7

DOI: 10.55124/jaim.v4i1.300 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Artificial intelligence and Machine Learning, 2024. Verfasst von Karthik Perikala.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Modern machine learning systems depend on large, continuously evolving feature sets to support both offlinetrainingandlow- latencyonlineinference. As feature ecosystems grow, bespoke pipelines and tightly coupled application logic becomeincreas- ingly difficult to operate, leading to inconsistent fea- ture definitions, operational fragility, and slow iter- ation cycles. This paper presents the design and evolution of a production feature serving platform, beginning with an in-house NoSQL- based Feature Mart and fol- lowed by a migration to a managed Feature Store. Thein-housesystemprovidedflexibleschemadesign and low-latency access but required substantial en- gineering effort to maintain ingestion pipelines, fea- ture registries, refresh workflows, and serving mi- croservices. The managed platform introduced automated in- gestion from analytical sources, built-in feature gov- ernance, and native serving interfaces, fundamen- tally shifting the operational model of feature lifecy- cle management.Using normalized empirical mea- surements, we evaluate latency percentiles, through- put stability, and execution behavior under live con- figuration updates. Our results show that managed feature stores can preserve predictable serving performance while sig- nificantly reducing operational complexity, making them a compelling foundation for large-scale pro- duction machine learning systems.

Zitieren

APA

Perikala, K. (2024). Feature Serving at Scale: Design, Migration, and Empirical Evaluation of a Production Feature Platform. Journal of Artificial intelligence and Machine Learning, 2(1), 1-7. https://doi.org/10.55124/jaim.v4i1.300

Vancouver

Perikala K. Feature Serving at Scale: Design, Migration, and Empirical Evaluation of a Production Feature Platform. Journal of Artificial intelligence and Machine Learning. 2024;2(1):1-7. doi:10.55124/jaim.v4i1.300

BibTeX
@article{Perikala2024,
  author = {Perikala, Karthik},
  title = {Feature Serving at Scale: Design, Migration, and Empirical Evaluation of a Production Feature Platform},
  journal = {Journal of Artificial intelligence and Machine Learning},
  year = {2024},
  volume = {2},
  number = {1},
  pages = {1--7},
  doi = {10.55124/jaim.v4i1.300}
}
Kostenlos starten →