R1-2410105
discussion
Additional study on other aspects of AI/ML model and data
From OPPO
Summary
OPPO's study document on AI/ML model identification and data management proposes 9 key technical proposals covering model identification procedures, lifecycle management approaches, and model transfer mechanisms for 3GPP Release 19 NR air interface AI/ML work item.
Position
OPPO advocates FOR a unified lifecycle management (LCM) framework that combines both functionality-based and model ID-based operations, supporting network-assigned model IDs with 1-to-1 mapping between model IDs and associated IDs. They push FOR Alternative B in model transfer procedures due to reduced signaling overhead and AGAINST studying partial parameter transfer and certain model identification options (MI-Option 5, MI-Option 2 for one-sided models) for Release 19 due to complexity and feasibility concerns.
Key proposals
- Proposal 1 (Model identification and LCM): Support a unified LCM providing both functionality-based and ID-based operations, with functionality-based operation as default aligned with Feature/FG granularity, and Model ID used on top for indicating different additional conditions
- Proposal 2 (Model identification and LCM): UE-side burden may be reduced if UE assumes NW-side additional conditions with same associated ID are consistent among multiple cells, but commonality between cells may be implied if associated ID ensures consistency
- Proposal 3 (MI-Option 1): Step D should be supported for UE not involved in Steps A/B/C, with Alt.1 (NW assigns Model ID) preferred as it supports identification for both involved and non-involved UEs
- Proposal 4 (MI-Option 2): ID-X is the Model ID or 1-to-1 mapped to Model IDs, with NW assigning Model ID in Step A and Step C needed if UE-part model used by UEs not involved in development
- Proposal 5 (Type A model identification): Model ID allocated to model and additional conditions used for training via OTT inter-vendor engineering
- Proposal 6 (Functionality and Model ID): Functionality ID can be used for indication of functionality between NW and UE
- Proposal 7 (Functionality and Model ID): Local model ID can be a simple number similar to resource/configuration ID in legacy NR specification without explicit model information
- Proposal 8 (Model transfer): Alt. B preferred for Case z4 due to smaller signaling overhead, using Model structure ID for candidate/supported lists and complete model ID for transferred parameters
- Proposal 9 (Model transfer): UE needs transfer of new parameters for known model structure when adapting to new scenario, with single report for dynamic AI/ML capability changes and deprioritize partial parameter transfer study