Dynamic Data Conversion Module for Privacy-Preserving Distributed ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Fragmented and private electronic health records across different medical institutes pose challenges for data sharing due to privacy concerns, making it difficult to coordinate distributed machine learning processes across multiple client devices.
Innovation Solution
A learning system deploys dynamic data conversion and learning modules to client devices, enabling them to collaborate in a distributed learning framework by performing ETL operations and training machine-learning models without exposing client data, using customized conversion logic and vocabularies to convert data into a common schema compatible with the master model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If healthcare data are shared across multiple medical institutes to train machine-learning models, then the model training effectiveness and data diversity are improved, but patient privacy and data security are compromised
Solution Approach 1:
The patent segments the machine-learning model into multiple components distributed across different medical institutes. Each institute trains local model segments on their own data without sharing raw patient records, thus improving model effectiveness through multi-institute collaboration while maintaining patient privacy through data localization
Solution Approach 2:
The patent introduces a central coordination server as an intermediary that manages model aggregation and distribution without accessing sensitive patient data. This mediator enables cross-institute collaboration by coordinating model training and updating while preventing direct data sharing between institutes
2Manufacturing precision
If ETL operations are manually configured at each client device to convert data to target schema, then data conversion accuracy is improved, but system complexity and deployment time increase
Solution Approach 1:
The patent implements self-service ETL operations where the system automatically discovers data schemas, generates conversion logic, and configures extraction, transformation, and loading operations without manual intervention. This maintains high conversion accuracy through automated schema validation while eliminating configuration complexity for users
Solution Approach 2:
The patent dynamically adjusts ETL parameters and conversion logic based on the specific characteristics of each client device's data schema. By automatically detecting and adapting to different data formats, structures, and vocabularies, the system achieves accurate data conversion across diverse sources without requiring manual configuration for each device
3Adaptability or versatility
If customized conversion logic is deployed to each client device to handle local data variations, then data compatibility with master model is improved, but deployment and maintenance complexity increase
Solution Approach 1:
The patent creates a universal conversion framework that handles diverse local data variations through a single deployable module. This ETL module can adapt to different data schemas, vocabularies, and structures across various client devices while maintaining compatibility with the master model, eliminating the need for separate customized deployments at each location
Data Source
AI summary
A learning system deploys a dynamic data conversion module (DDCM) that is customized to perform one or more extract, transform, and load (ETL) operations on the local client data that is used to train at least a portion of a master machine-learning model for the distributed learning framework. The DDCM envelopes a set of components including at least a pre-assessment toolkit and a ETL model. The pre-assessment toolkit is configured to collect statistics and abstract information from the client database of the client device and provide the statistics and abstract information to the learning system. Based on the statistics and abstract information of a respective client device, the learning system generates conversion logics and standardized vocabularies and provides them to the DDCM of the client device.