Terminal AI Model Refinement Using Shared-Layer Intermediate Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The risk of personal information leakage during the transmission of data to refine AI models stored in terminals is a concern due to the need for external communication, which compromises data processing speed and user privacy.
Innovation Solution
A method and system for refining AI models within terminals using intermediate data from shared neural network layers, processed by a server with a larger model, to enhance model performance without transmitting sensitive user data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If operation data is transmitted from terminal to server for model refinement, then model performance is improved, but user personal information security deteriorates
Solution Approach 1:
The patent extracts only the necessary intermediate data from the terminal's model processing without transmitting the complete operation data containing personal information. By taking out only the essential features needed for model refinement while leaving sensitive data in the terminal, the system achieves model improvement without exposing user privacy.
Solution Approach 2:
The patent introduces an intermediary mechanism where intermediate data serves as a mediator between the terminal and server. This intermediate representation allows the server to refine models without directly accessing sensitive operation data, thus protecting personal information while still enabling model performance improvement.
2Measurement precision
If complete operation data is transmitted to server, then model refinement accuracy is improved, but data transmission security and processing efficiency deteriorate
Solution Approach 1:
The patent extracts only the essential intermediate data needed for model refinement from the complete operation data. This extraction process removes unnecessary and sensitive information, reducing the data volume transmitted to the server while maintaining sufficient information for accurate model refinement, thus improving processing efficiency without sacrificing refinement quality.
3Object-affected harmful factors
If intermediate data is transmitted instead of complete data, then user privacy protection is improved, but model refinement effectiveness may deteriorate
Solution Approach 1:
The patent applies local quality by transmitting intermediate data with specific properties tailored for model refinement. The intermediate data maintains the essential features and patterns needed for effective model learning while having different characteristics from the original sensitive operation data, thus achieving both privacy protection and refinement effectiveness.
Data Source
AI summary
A method performed by a server is provided. The method includes, in response to input information being processed by the first model in the terminal, receiving, from the terminal, intermediate data output from a first layer included in a shared portion of the first model, obtaining the first model and a second model, which are pretrained by including the shared portion, obtaining correct answer data for the input information by inputting the intermediate data as an output of the first layer to the first layer included in the shared portion of the second model, in response to the intermediate data being input as the output of the first layer to the first layer included in the shared portion of the first model, refining the first model so that the correct answer data may be output from the first model, and transmitting information about the refined first model to the terminal.