R1-2410776
discussion
Summary #2 for other aspects of AI/ML model and data
From OPPO
Summary
This OPPO-moderated document presents a comprehensive summary of AI/ML model identification, training data collection, and model transfer/delivery discussions for RAN1 #119, containing over 40 proposals across multiple technical areas. The document addresses three main topics: model identification procedures for two-sided models (MI-Options 1-4), training data collection mechanisms for UE-sided models, and model transfer/delivery procedures (particularly Case z4).
Position
OPPO, as document moderator, advocates for a balanced approach supporting both functionality-based and model ID-based LCM operations while prioritizing standardized model structures over offline vendor collaboration. They push FOR unified LCM frameworks, 3GPP-specified model structures for Case z4, and network-assigned model IDs to reduce cross-vendor coordination burden. They position AGAINST offline alignment approaches and overly complex multi-ID schemes, favoring implementation transparency and reduced proprietary information disclosure.
Key proposals
- Proposal 2.1 (Sec 2.1): Regarding MI-Option2 for two-sided models, investigate three alternatives for UE-part identification: Alt. MI-2A using corresponding ID-X, Alt. MI-2B using network-assigned model ID, and Alt. MI-2C using UE-reported model ID
- Proposal 2.2 (Sec 2.2): For MI-Option2 dataset transfer, transmit input data/labels for UE part, format/type information, dataset size, and validation/testing information from network to UE
- Proposal 2.6 (Sec 2.6): Study MI-Option4 reference models with three cases: standardized UE part only (MI-4A), NW part only (MI-4B), or both parts (MI-4C) of two-sided models
- Proposal 4.1 (Sec 4.1): Observation on aligning known model structures for Case z4: 3GPP specification reduces cross-vendor burden but requires large standardization effort, while offline alignment reduces spec effort but increases collaboration burden
- Proposal 4.2 (Sec 4.2): Focus RAN1 study on standardized known model structures (Opt.1) for Case z4, as offline alignment is beyond RAN1 expertise scope
- Proposal 4.4 (Sec 4.4): Study three open format options for Case z4: reuse existing formats like ONNX, define new 3GPP format using ASN.1, or reuse SA2 interoperability token mechanism
- Proposal 4.5 (Sec 4.5): For Case z4 model readiness determination, consider UE signaling notification, minimum application time specification, or UE-reported minimum time with network configuration options
- Proposal 4.8 (Sec 4.8): Consider transformer, CNN, and LSTM as standardized model structure backbones for Case z4 inference as starting points
- Proposal 2.1A (Summary): For MI-Option2 study, ID-X can be used for pairing UE-part and NW-part of two-sided models with further study needed on additional pairing information
- Proposal 4.7 (Summary): Study relationship between model ID, first indication, and second indication with three options: combined information, second indication as model ID, or network-assigned separate model ID
- Proposal 2.5 (Summary): Conclude that whether dataset includes single or multiple cell data is network implementation choice and transparent to UE/UE-side
- Proposal 4.5B (Summary): Identify candidate solutions for Case z4 model readiness including UE signaling notification and minimum application time assumptions with UE specification/reporting