Learning Assistance Device for Confidential Medical Model Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In the medical field, collecting and efficiently utilizing large amounts of high-quality learning data for deep learning is challenging due to data confidentiality and the high cost of data collection, while maintaining the quality of artificial intelligence requires a mechanism for evaluating performance after learning.
Innovation Solution
A learning assistance device and system that acquires and evaluates learned discriminators from multiple terminal devices, selects the discriminator with the highest correct answer rate, and updates the model using different image data, ensuring confidentiality by not distributing image data externally, thereby improving discrimination performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If image data is collected from multiple medical institutions for deep learning, then the quantity and variety of learning data is improved, but data confidentiality and security are compromised
Solution Approach 1:
The system segments the learning data collection process by having each terminal device learn using only its own local image data. The discriminator is divided into a learning discriminator (for training with local data) and an actually operated discriminator (for deployment). This segmentation allows data to remain localized at each institution, preventing consolidation while still enabling distributed learning accumulation.
Solution Approach 2:
The learning assistance device acts as an intermediary that collects not the raw image data itself, but rather the learned discriminators (model parameters) from multiple terminal devices. This intermediary approach allows the system to aggregate learning results without aggregating sensitive patient data, maintaining confidentiality while achieving the benefits of diverse learning data.
2Reliability
If learning data is collected from multiple sources, then the quality of deep learning model is improved, but the cost and complexity of data collection increases
Solution Approach 1:
Each terminal device performs self-service learning by automatically using its own local image data to train the learning discriminator. The device independently manages its own learning process without requiring external data collection infrastructure, reducing complexity while still contributing to the overall system improvement through the aggregation of learned discriminators.
Solution Approach 2:
Instead of collecting and centralizing original image data, the system collects copies in the form of learned discriminators (model parameters) from each terminal device. These discriminator copies encapsulate the learning results without containing the sensitive source data, achieving model quality improvement through aggregation while avoiding the complexity of secure data collection and management.
3Ease of operation
If the same learning discriminator is used across multiple terminal devices, then ease of deployment is improved, but discrimination performance is limited by local data variety
Solution Approach 1:
The system implements dynamic model evolution where the learning discriminator is continuously updated at each terminal device using local data, and the best performing learned discriminators are aggregated and redistributed. This dynamic process allows the model to adapt to local conditions while progressively improving overall performance through accumulated learning from diverse data sources across multiple devices.
Solution Approach 2:
The system establishes a feedback loop where learned discriminators from multiple terminal devices are collected, evaluated, and the best results are fed back to update the learning discriminator for the next learning cycle. This feedback mechanism ensures continuous performance improvement while maintaining ease of deployment, as each device receives updated models through the learning assistance device without requiring complex manual intervention.
Data Source
AI summary
The learning assistance device acquires a plurality of learned discriminators obtained by causing learning discriminators provided in a plurality of respective terminal devices to perform learning using image correct answer data, sets the learned discriminator having the highest correct answer rate among the plurality of learned discriminators as a new learning discriminator, and outputs the learning discriminator and identification information capable of identifying the image correct answer data used for learning. The plurality of terminal devices repeatedly performs a process of outputting a plurality of learned discriminators obtained by causing the learning discriminators to perform learning using image correct answer data different from image correct answer data indicated by the identification information.