R1-2409503
discussion
Discussion on other aspects of AI/ML model and data
From CMCC
Summary
CMCC discusses model identification options (MI-Option 2, 3, and 4) and model transfer/delivery cases for NR AI/ML, proposing to prioritize MI-Option 2 and 3 while deprioritizing MI-Option 4 due to specification effort and RAN4 dependencies. The document outlines specific procedures for dataset transfer, model parameter exchange in Case z4, and the relationship between model IDs and indications, totaling 15 proposals.
Position
CMCC proposes prioritizing MI-Option 2 and MI-Option 3 for model identification while deprioritizing MI-Option 4 due to specification efforts and RAN4 dependencies. For MI-Option 2, they propose studying whether the NW or UE assigns the model ID and suggest that dataset IDs can serve as ID-X. They argue that meta information, including CSI generation/reconstruction I/O and model backbone, is necessary for dataset transfer validity. Regarding model transfer/delivery Case z4, CMCC proposes standardizing reference model structures and exchanging parameters, with specific procedures for UE capability reporting and parameter updates based on performance requirements. They present three alternatives for the relationship between Model ID, first indication (structure), and second indication (parameters), and propose that partial parameters share the same second indication. Finally, they suggest using existing RAN1 Reply LS as the baseline for UE data collection mechanisms.
Key proposals
- Proposal 1 (Model information during model identification): Proposes that model information should include related functionality/AI enabled feature, applicable scenarios/configurations, and type/dimension of model input/output.
- Proposal 2 (Procedure of MI-Option 2): Suggests that a dataset ID associated with the transferred dataset in Step A of MI-Option 2 can be interpreted as a kind of ID-X.
- Proposal 3 (Procedure of MI-Option 2): Suggests further studying two alternatives for model ID determination in MI-Option 2: NW assigns model ID or UE assigns/reports model ID.
- Proposal 4 (Procedure of MI-Option 2): Proposes that meta information such as CSI generation/reconstruction I/O, data sample format, scalability, quantization method, and model backbone may be needed before/during dataset transfer.
- Proposal 5 (Procedure of MI-Option 3): Proposes a three-step procedure for MI-Option 3: NW transmits model list, UE reports supported list, and NW transfers model and assigns ID.
- Proposal 6 (Procedure of MI-Option 4): Proposes deprioritizing MI-Option 4 at the current stage due to limited performance gain, high specification efforts, and ongoing RAN4 studies.
- Proposal 7 (Summary of model identification options): Suggests prioritizing MI-Option 2 and MI-Option 3 for further study.
- Proposal 8 (Applicable scenarios): Proposes that model transfer/delivery usages include deployment for one-sided/two-sided models, pairing for two-sided models, and ensuring NW-side additional conditions consistency.
- Proposal 9 (Discussion on model transfer/delivery case z4): Suggests further studying Case z4 regarding standardizing reference model structure, exchanging model parameters, and associated procedures.
- Proposal 10 (Discussion on model transfer/delivery case z4): Proposes adding Step A-0 for UE capability reporting on support for Case z4 before reporting supported known model structures.
- Proposal 11 (Discussion on model transfer/delivery case z4): States that transferring new parameters for a known model structure may be needed after offline engineering or if previous parameters fail performance requirements.
- Proposal 12 (Discussion on model transfer/delivery case z4): States that for Option 3a/5a, the AI model with transferred parameters is ready for inference if it meets performance requirements and UE implementation factors.
- Proposal 13 (Discussion on model transfer/delivery case z4): Proposes studying three alternatives for the relationship between Model ID, first indication (structure), and second indication (parameters).
- Proposal 14 (Discussion on model transfer/delivery case z4): Proposes that for partial parameter transfer, each partial parameter should be associated with the same second indication.
- Proposal 15 (UE data collection): Suggests RAN2 take the Reply LS on Data Collection Requirements and Assumptions (R1-2310681) as the baseline for UE side data collection mechanisms.