R1-2410196
discussion
Discussion on other aspects of AI/ML model and data
From LG Electronics
Summary
LG Electronics presents 8 proposals addressing AI/ML model identification, lifecycle management, and transfer mechanisms for NR air interface. The document focuses on clarifying functionality granularity, model identification options, and addressing practical challenges in model delivery scenarios.
Position
LG Electronics advocates for a functionality-based approach to AI/ML model lifecycle management with flexible sub-use-case specific configurations, while pushing against overly rigid model identification schemes. They emphasize practical implementation challenges like cross-vendor collaboration and proprietary design disclosure, advocating for performance monitoring-based approaches rather than complex identification mechanisms.
Key proposals
- Proposal 1 (Model/functionality identification): Clarify that a functionality is a unit that affects NW-side and/or UE-side operation/performance, with granularity same/similar to an FG and can be smaller than a sub-use-case
- Proposal 2 (Model/functionality identification): Clarify that NW can configure NW-side additional condition to UE for helping UE select applicable functionality, with exact configuration method including associated ID being up to sub-use-case signaling design
- Proposal 3 (Model/functionality identification): Details of associated ID, including whether needed, name, inferred information, and assignment/usage should be discussed per sub-use-case
- Proposal 4 (Model/functionality identification): For MI-Option 4, procedures like performance/compatibility report to ensure model performance need consideration
- Proposal 5 (Model/functionality identification): For MI-Option 2, clarify that ID-X alone cannot serve model identification purpose and discuss if model identification is necessary for dataset transfer scenario
- Proposal 6 (Model delivery/transfer): Focus on discussing key challenges like offline cross-vendor collaboration, model storage requirements, and proprietary design disclosure issues
- Proposal 7 (Model delivery/transfer): Whether/when UE needs transfer of new parameters for known model structure can be known by NW via performance monitoring procedure
- Proposal 8 (Model delivery/transfer): Whether/when AI model with transferred parameters is ready for inference can be known by NW via defining model application time as minimum required time for UE to apply transferred model