Model Sub-Model Segmentation for Trusted Execution Environment Security

The patent divides the complete model into multiple sub-models and selectively deploys only necessary sub-models to terminal devices. This segmentation approach allows the system to maintain model security by keeping sensitive parts of the model on secure servers while still enabling local processing of specific tasks, thus resolving the contradiction between processing efficiency and model security.

The patent applies different quality levels to different parts of the model by identifying and selecting specific sub-models with appropriate security attributes for local deployment. Critical model components remain on secure servers while less sensitive sub-models are deployed to terminals, creating a differentiated security architecture that balances efficiency and security.

The patent performs preliminary model sub-setting and attribute parameter analysis before actual deployment. By pre-processing the model into sub-models with defined attribute parameters and pre-establishing the correspondence between sub-models and terminal device capabilities, the system reduces runtime complexity while maintaining high adaptability to different devices.

The patent utilizes attribute parameters (such as security levels, computational requirements, memory constraints) to dynamically select and configure appropriate sub-models for different terminal devices. This parameter-based selection mechanism enables flexible adaptation to diverse devices without requiring complex manual configuration, as the system automatically matches device capabilities with suitable sub-model parameters.