|
Work item:
|
Y.paid-reqts
|
|
Subject/title:
|
Data handling - Functional requirements of AI data pipeline
|
|
Status:
|
Under study
|
|
Approval process:
|
AAP
|
|
Type of work item:
|
Recommendation
|
|
Version:
|
New
|
|
Equivalent number:
|
-
|
|
Timing:
|
2028-Q2 (High priority)
|
|
Liaison:
|
ITU-T SG 17, SG20, SG21, ISO/IEC JTC 1/SC 42
|
|
Supporting members:
|
ETRI, Zhejiang Lab, Kyung Hee Univ, Korea (Republic of)
|
|
Summary:
|
As AI systems are deployed across domains including healthcare, financial services, and industrial manufacturing, producing data that is directly and reliably usable for model training, validation, and inference has become a critical operational requirement. Such data for AI is produced consistently through structured pipeline operations covering data collection, data preparation, transformation, verification, and provision.
Standardized functional requirements of pipeline for data for AI do not yet exist. In particular, AI data stage definitions, stage-level input and output specifications, and the binding between pipeline execution instances and produced dataset versions are not addressed in current standards. This gap impairs both operational efficiency, through repetitive and non-interoperable data preparation work, and trustworthiness, through the absence of standardized audit records and traceability linkages.
This work item studies the functional requirements necessary to support consistent, interoperable, and reproducible production of data for AI, including production of the structured outputs. The resulting Recommendation is intended to provide a common functional requirements for pipeline providers operating pipeline for data for AI.
|
|
Comment:
|
-
|
|
Reference(s):
|
|
|
Historic references:
|
|
Contact(s):
|
|
| ITU-T A.5 justification(s): |
|
|
|
|
First registration in the WP:
2026-07-23 14:24:42
|
|
Last update:
2026-07-23 14:30:20
|
|
