R1-2407800
discussion
Discussion on other aspects of AI/ML model and data
From ZTE
Summary
ZTE's contribution analyzes AI/ML model identification and transfer options for NR air interface, comparing dataset transfer (MI-Option 2), model transfer (MI-Option 3), and standardized reference models (MI-Option 4). The document contains 12 proposals and 5 observations across model identification, model transfer/delivery, and data collection aspects.
Position
ZTE strongly advocates AGAINST MI-Option 2 (dataset transfer) due to feasibility concerns including huge resource overhead (1-10GB datasets vs 7-100MB models), large latency, and high UE power consumption. They strongly support MI-Option 3 (model transfer) and MI-Option 4 (standardized reference models) as more practical alternatives. ZTE pushes FOR prioritizing model transfer case z4 with known model structures and Type B model identification over Type A, while advocating for two-sided models over UE-side models for Rel-19 normative work.
Key proposals
- Proposal 1 (Sec 2.1.1): Regarding MI-Option2, dataset ID is considered as model ID
- Proposal 2 (Sec 2.1.1): Regarding MI-Option 2, RAN1 further studies the necessity and potential approaches to deal with the impact of UE-side additional conditions for the dataset
- Proposal 4 (Sec 2.1.3): Regarding MI-Option 4, standardization of reference UE-part model is preferred
- Proposal 6 (Sec 2.1.4): In Rel-19 AI/ML framework study, type B model identification is prioritized compared with type A model identification
- Proposal 7 (Sec 2.2): In Rel-19 AI/ML framework study, RAN1 prioritizes the model transfer study for two-sided model rather than UE-side model
- Proposal 8 (Sec 2.2): In Rel-19 AI/ML framework study, RAN1 prioritizes the model transfer z4 for two-sided model
- Proposal 9 (Sec 2.2): The details of Alt. A and Alt. B of model transfer can be further clarified in normative work phase, e.g., information exchange about the buffered parameters at the UE side
- Proposal 10 (Sec 2.2): The details of specification impacts of model structure can be further studied and discussed in CSI compression agenda item
- Proposal 11 (Sec 2.2): Regarding model transfer/delivery, the details of signalling related specification impacts can be further studied in RAN2 working group
- Proposal 12 (Sec 2.3): Regarding CN/OAM/OTT collection of UE-sided model training data, RAN1's work can be triggered by RAN2 LS if needed
- Proposal 3 (Sec 2.1.1): Regarding MI-Option 2, dataset transfer can be realized by higher layer signalling and the detailed mechanism of dataset transfer is up to RAN2
- Proposal 5 (Sec 2.1.3): Regarding MI-Option 4, standardization of reference model can be further studied in CSI compression agenda item to avoid duplicated discussion