R1-2409753
discussion
Discussion on Other aspects of AI/ML model and data
Summary
Tejas Networks discusses model identification and data handling for AI/ML in NR, focusing on the consistency of NW-side additional conditions via 'associated IDs' and the mapping between these IDs, datasets, and model IDs. The document contains 7 observations and 5 proposals covering training/inference consistency, ID computation, and model transfer mechanisms.
Position
Tejas Networks proposes that use-case specific data collection configurations and site-specific information be mapped to associated IDs to ensure consistency between training and inference. They claim the UE should assign model IDs and report them to the NW, rather than the NW assigning them, due to the UE's awareness of trained models. They propose computing the associated ID based on PLMN ID and legacy RRC message IDs. They support flexible mappings (one-to-one, many-to-one, etc.) between associated IDs and model IDs, specifically highlighting many-to-one mappings for generalized models. They propose that a data set ID relates to multiple associated IDs for two-sided models. They note that model IDs correspond one-to-one with known model structures in transfer scenarios.
Key proposals
- Proposal 1 (Handling of NW-side additional conditions): Proposes that use-case specific data collection-related configurations and area or site-specific information are mapped to associated ID(s).
- Proposal 2 (Handling of NW-side additional conditions): Claims that the UE should assign a model ID and report it back to the Network (NW).
- Proposal 3 (Mapping between associated ID(s) to model ID(s)): Proposes that in a many-to-one mapping, a single model is trained by the UE after collecting many associated ID(s) corresponding to different NW side configurations.
- Proposal 4 (Mapping between associated ID(s) to model ID(s)): Proposes that the associated ID is computed based on PLMN ID and other IDs existing in legacy RRC message structures per use case.
- Proposal 5 (Model identification with data set transfer): Proposes that a data set ID is related to multiple associated ID(s).
- Observation 1 (Handling of NW-side additional conditions): Notes that NW side additional conditions are mapped to associated ID(s).
- Observation 2 (Handling of NW-side additional conditions): Notes a one-to-one mapping between associated ID(s) to data set(s) or data set ID(s).
- Observation 3 (Mapping between associated ID(s) to model ID(s)): Notes that mapping can be one-to-one, one-to-many, many-to-one, or many-to-many.
- Observation 4 (Mapping between associated ID(s) to model ID(s)): Notes that multiple associated ID(s) may correspond to a single data set, allowing for generalized model training.
- Observation 5 (Model identification with data set transfer): Notes that NW transmits data set ID(s) along with data sets for the UE part of two-sided models.
- Observation 6 (Model transfer/delivery): Notes a one-to-one correspondence between model ID and the first indication referring to known model structure.
- Observation 7 (Model transfer/delivery): Notes that for Neural Networks, transmitted parameters include layers, weights, biases, and activation functions.