R1-2410378
discussion
Discussion on AI/ML for CSI prediction
From NTT DOCOMO
Summary
NTT DOCOMO proposes criteria for deciding whether to standardize training/inference consistency mechanisms for AI/ML CSI prediction, based on comparing performance losses to the marginal 0-7.8% UPT gains observed in Release 19 studies. The document contains 2 proposals focused on reusing beam management mechanisms as a baseline solution.
Position
NTT DOCOMO advocates FOR a pragmatic approach to AI/ML CSI prediction standardization, arguing that consistency mechanisms should only be standardized if performance losses are significant relative to the already marginal gains. They push FOR reusing existing beam management mechanisms rather than developing new solutions, emphasizing practical deployment considerations over theoretical performance optimizations.
Key proposals
- Proposal 1 (Sec 2.1): Decide whether to support the consistency of training/inference according to the relative performance loss to the identified performance gain from the Rel. 19 study. Support the normative work of the consistency if the performance loss is non-ignorable (e.g., in the same order of magnitude) compared with the AI/ML prediction gain over non-AI/ML benchmark (e.g., 0-7.8% mean UPT gain or corresponding SGCS gain).
- Proposal 2 (Sec 2.2): The mechanism of AI/ML for beam management can be reused as a baseline to ensure the consistency of AI/ML CSI prediction. FFS: The necessity of further enhancements.