Abstract
Cloud-based machine learning (ML) platforms enable ML practitioners to build, rebuild, and serve multiple machine learning models in production environments. Proper ML metadata tracking and management is important to enable large-scale experimentation and to provide traceability and verifiability for modern production ML. This disclosure describes a ML metadata service to manage the lifecycle of metadata consumed and produced by ML pipelines. The ML metadata service enables logging detailed metadata as artifacts, capturing metadata as typed artifacts, and capturing a ML pipeline in an intuitive workflow graph. The metadata service enables provision of a ML dashboard that displays visualizations of a ML workflow along with the relevant metadata for each type of entity and supports queries for models and/or datasets that meet specific criteria.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Alfred, Ajay; Dournov, Pavel; Wu, Jingxiao; Sun, Hongye; Verma (AP), Apoorv; Gopinathan, Ajay; Miao, Hui; and Polyzotis, Neoklis, "Metadata Service to Enable Display of Rich Artifacts in Machine Learning Pipelines", Technical Disclosure Commons, (December 04, 2020)
https://www.tdcommons.org/dpubs_series/3850