R1-2409929
discussion
Additional study on AI/ML for other aspects
From CATT
Summary
This 3GPP RAN1 technical document from CATT addresses various aspects of AI/ML for NR air interface, focusing on model identification, model transfer/delivery, and data collection issues. The document contains 15 proposals and 3 observations covering model identification options, transfer procedures, and feasibility assessments for two-sided AI/ML models.
Position
CATT advocates for flexible and practical approaches to AI/ML model management, pushing FOR simplified model identification schemes with flexible dataset-to-model mappings while pushing AGAINST overly complex standardized model transfer mechanisms. They specifically argue that model transfer Case z4 is not necessary for Rel-19 due to feasibility challenges and availability of alternative solutions, and advocate for focusing study efforts on two-sided models only while deprioritizing UE-first training approaches.
Key proposals
- Proposal 1 (Sec 2.1): No need to discuss MI-Option 1 or associated ID in this agenda since it's out of scope for two-sided models
- Proposal 2 (Sec 2.2.1): 'ID-X' in MI-Option2 can at least be used for identifying a dataset constructed by certain data samples, with FFS other usage
- Proposal 4 (Sec 2.2.2): In AI-Example2-1, mapping relationship between dataset ID (ID-X) and model ID is flexible, not limited to one-on-one mapping
- Proposal 5 (Sec 2.2.3): In AI-Example2-1, for AI/ML-based CSI compression, the need and benefit of UE-side additional condition(s) is unclear
- Proposal 6 (Sec 2.2.4): No need to open discussion on UE-first training/dataset sharing to coincide with CSI compression study
- Proposal 7 (Sec 3.1): For Alt.A of model transfer case z4, add Step A-0 where NW requests UE to report supported known model structure
- Proposal 9 (Sec 3.1): Study two cases for new parameter transmission - UE-initiated (cell changes) and NW-initiated (model updates)
- Proposal 10 (Sec 3.1): Two options for model readiness - confirmative message from UE or predefined maximum gap/delay assumption
- Proposal 12 (Sec 3.2): Three directions for aligning model structure understanding - standardized reference model + model ID, standardized structure + structure ID, or model description format
- Proposal 14 (Sec 3.3): Model transfer Case z4 is not necessary (alternatives exist), has feasibility challenges, but provides benefits that other alternatives can also achieve
- Proposal 15 (Sec 4): Discussion on CN/OAM/OTT collection of UE-sided model training data is left to RAN2 and/or RANP based on SA feedback