Distributed Medical Inference Model Segmentation for Privacy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical inference processing systems face challenges in securing the confidentiality of inference models and protecting the privacy of medical data, as they often require transmitting sensitive information to external servers, making it difficult to prevent model abuse and data privacy breaches.
Innovation Solution
The system employs a dual-apparatus architecture where a first information processing apparatus performs partial inference processing using a first partial model, and a second apparatus performs subsequent processing using multiple partial models, ensuring confidentiality by only sharing results and selection information, thus reducing data transmission and resource usage while maintaining model security.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the complete inference model is deployed in a single information processing apparatus, then inference processing can be performed independently without external communication, but the confidentiality of the inference model cannot be secured as users may copy or alter the model
Solution Approach 1:
The inference model is divided into multiple partial models (first partial model and second partial model) that are distributed across different information processing apparatuses. The first apparatus holds the input layer and some intermediate layers, while the second apparatus holds the remaining intermediate layers and output layer. This segmentation prevents any single apparatus from possessing the complete model, thereby securing confidentiality while enabling distributed inference processing.
2Adaptability or versatility
If medical data is transmitted to external cloud services for inference processing, then advanced inference capabilities can be utilized, but the privacy of the medical data cannot be protected
Solution Approach 1:
The inference model is segmented and distributed across multiple apparatuses, with the first apparatus performing initial inference processing locally using the first partial model, and only transmitting intermediate results (not raw medical data) to the second apparatus. This segmentation allows utilization of advanced inference capabilities while protecting medical data privacy by keeping sensitive data local and transmitting only processed intermediate representations.
Solution Approach 2:
The first inference result serves as an intermediary representation between the raw medical data and the final inference output. Instead of transmitting raw medical data to external services, the system transmits this intermediate result which contains less sensitive information, thereby enabling external inference processing while protecting patient privacy.
3Reliability
If the inference model is kept centralized for security, then model confidentiality is maintained, but data transmission to external services is required which compromises privacy
Solution Approach 1:
The model is segmented and distributed such that no single external service receives the complete model or raw data. The first apparatus retains the first partial model locally, performing initial processing without external communication for the most sensitive data, thereby maintaining model confidentiality while minimizing data transmission risks.
Data Source
AI summary
An information processing system includes a first information processing apparatus including a first inference unit configured to perform first inference processing on inference target medical data using a first partial model including an input layer and at least some of intermediate layers and corresponding to a plurality of second partial models, and a first output unit configured to output a result of the first inference processing and selection information to a second information processing apparatus, and the second information processing apparatus including a second inference unit configured to perform second inference processing by inputting a result of the first inference processing to a second partial model selected from among the plurality of second partial models based on the selection information.