R1-2410052
discussion
Discussion on other aspects of AI/ML model and data
From Fujitsu
Summary
Fujitsu's contribution addresses unresolved aspects of AI/ML model identification and model transfer/delivery for NR air interface, presenting 12 proposals and 1 observation across model identification procedures (MI-Option2), model transfer cases, and unified procedures for two-sided models.
Position
Fujitsu advocates FOR comprehensive dataset-based model identification (MI-Option2) with proper ID management and test dataset procedures, while pushing AGAINST impractical model transfer cases - specifically arguing to deprioritize Case z1 (due to unclear benefits over Case y) and Case z4 for UE-sided models (due to real-time update impracticality), and opposing partial parameter transfer studies due to complexity and lack of justification.
Key proposals
- Proposal 1 (Sec 2.1): At least the ID of the NW part model for generating the transferred dataset(s) should be provided to UE during dataset transfer
- Proposal 4 (Sec 2.1): Test dataset for UE part model development should be studied including how to generate test dataset, distinguish from training dataset, performance assessment, and completion of model development
- Proposal 6 (Sec 2.4): A unified model identification procedure after model update based on either dataset transfer or model transfer should be studied for both MI-Option 2 and MI-Option 3
- Proposal 7 (Sec 3): Deprioritize Case z1 if its benefit over Case y from the location of model storage cannot be justified
- Proposal 8 (Sec 3): Model transfer/delivery Case z4 is deprioritized for UE-sided model in Rel-19 from RAN1 perspective
- Proposal 9 (Sec 3): Model ID should be indicated by combination of first indication (known model structure) and second indication (model parameters)
- Proposal 10 (Sec 3): Both UE-initiated and NW-initiated transfer can be assumed when version number at NW is higher than at UE or parameter differences are identified
- Proposal 11 (Sec 3): Performance of AI model with transferred parameters should be tested before activation - ready if performance is better than before update or meets performance request
- Proposal 12 (Sec 3): Study on transfer of partial parameters for known model structure should be deprioritized due to lack of evaluation results and additional inter-vendor collaboration workload
- Proposal 2 (Sec 2.1): ID of dataset is assumed for purposes of model pairing or model identification
- Proposal 3 (Sec 2.1): Additional ID to identify different transferred datasets generated from the same NW part model should be supported
- Proposal 5 (Sec 2.1): ID of NW part model for generating transferred dataset(s) should be reported to NW for model identification or model pairing