Federated Intervention Prediction with Cross-Organization Feature Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques struggle to predict the effect of an intervention on a subject when the organization acquiring the intervention situation and the organization acquiring the result are different, and the features available to each are not aligned.
Innovation Solution
Implement federated learning to train prediction models using converted features that align the feature distributions in a shared space across multiple information processing devices, allowing each device to predict the intervention effect without sharing raw data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If federated learning is used to train prediction models across multiple organizations, then data privacy and security are improved, but feature distribution alignment and model training effectiveness deteriorate
Solution Approach 1:
The patent introduces a feature space alignment mechanism as an intermediary between federated learning organizations. Each organization converts its local features to a shared feature space using conversion models, enabling effective model training while maintaining data privacy. The alignment process acts as a mediator that bridges the gap between heterogeneous feature distributions without requiring raw data sharing.
Solution Approach 2:
The patent transforms features by changing their parameter representation through conversion models. Local features from different organizations are converted to a standardized feature space with aligned distributions, allowing the federated learning model to effectively process heterogeneous data while preserving the privacy benefits of federated learning.
2Measurement precision
If features are converted to align distributions in shared space, then prediction model effectiveness is improved, but computational complexity and processing overhead increase
Solution Approach 1:
The patent performs feature space alignment as a preliminary action before federated model training. By pre-converting local features to the shared feature space and aligning their distributions in advance, the system reduces the computational burden during the actual federated learning process, making the overall system more efficient despite the added preprocessing step.
3Reliability
If organizations share minimal data through federated learning, then data security is improved, but information completeness and model training quality deteriorate
Solution Approach 1:
The patent creates a universal feature space that can accommodate features from multiple different organizations with different schemas and distributions. This multi-functional feature space enables the federated learning model to effectively utilize information from diverse sources while maintaining data security, as the alignment process preserves essential information patterns without requiring raw data sharing.
Data Source
AI summary
A first information processing device includes a first acquisition unit acquiring a first feature representing a subject and a result, a first prediction unit predicting an intervention situation that could have affected the result based on the first feature, and a first prediction model training unit training a prediction model that predicts an effect of intervention by federated learning based on a third feature converted from the first feature, the result, and the intervention situation. A second information processing device includes a second acquisition unit acquiring a second feature representing a subject and an intervention situation, a second prediction unit predicting a result based on the second feature, and a second prediction model training unit training a prediction model by federated learning based on the third feature converted from the second feature, the result, and the intervention situation.