Method of advanced model adaptation for radio access network
AssigneeAUMOVIO Germany GmbH
InventorsHojin KIM, Rikin SHAH
Life of this patent
Estimating the term…
Abstract
The present disclosure describes methods of model adaptation/switching for AI/ML based model in wireless mobile communication system including base station (e.g., gNB) and mobile station (e.g., UE). Threshold-based mapping relationship is configured to enable model adaptation using triggering information depending on dynamic changes of model operation environment at device side.
Claims9
A method of advanced model adaptation for radio access network by generating the pre-configured mapping relationship between applicable sub-model IDs with different threshold values, comprising:
- Threshold values are pre-defined to match with activation triggering of different AI/ML sub-models;
- Sub-model IDs are split from the common model ID, where sub-models can have lower model complexity/size compared with common model ID;
- The attribute data for threshold value calculation, {UE device resource specification, model configuration, site/scenario, application}, can be defined for different use case/application and implementation-specific environment together with considering UE capability;
- The listed sub-models can be dynamically configured with the pre-defined parameter set change information sent from network where the determined model ID can be macro model and the listed multiple sub-models can be micro models so that macro model is full featured model with highest complexity/size while micro model is partially featured model with lower complexity/size;
- Depending on practical use case, the determined model ID itself can be one of the indexed sub-model;
- The size of mapping table or number of sub-models can be adapted to the specific model operation applications and/or environment.
A method of forming full model (fML) and partial models (pML) for radio access network, comprising:
- Network generates two model categories such that fML is the original model having full feature set and pML is the simplified model having lower complexity and/or smaller feature set for applying any specific model(s) to UE;
- fML and pML are pre-configured so that model complexity is lower and feature set size is smaller for partial models where model configuration parameters such as number of layers and feature input can be adjusted to determine a finite set of full-/partial-models;
- Selection of models from fML and pMLs depends on UE ML capacity status to match with target model operation so that any pMLs can be run on the reduced available resource at UE device;
- Scalable model structure supports models with different complexity levels by using parameter configuration.
Dependencies are inferred from the claim text.