Privacy-Preserving Feature Aggregation for Detection Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for updating analytical models lack efficient mechanisms for rapid distribution and implementation across multiple local nodes while ensuring data privacy, especially in scenarios where sensitive information is shared across entities.
Innovation Solution
A computer-implemented method and system that aggregates data into features, generates detection models using these features, and transmits packages with instructions for calculating selected features and thresholds to local nodes for implementation, ensuring data protection through privacy management and tracking mechanisms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data is shared across multiple entity systems for model generation, then the robustness and accuracy of detection models is improved, but data privacy and security are compromised
Solution Approach 1:
The patent extracts only the necessary feature information from raw data for model training, leaving the sensitive underlying data behind. Local nodes aggregate data into features that describe content without exposing actual sensitive records, thereby improving model robustness while protecting data privacy.
Solution Approach 2:
The patent introduces an intermediary feature aggregation layer between raw sensitive data and the detection model. This intermediary transforms data into feature representations that preserve analytical value for model training while removing direct links to sensitive information, resolving the contradiction between data sharing and privacy protection.
2Productivity
If detection models are updated rapidly across multiple local nodes, then the system's responsiveness to new fraud patterns is improved, but data distribution and implementation complexity increases
Solution Approach 1:
The patent segments the model update process into independent components that can be distributed to multiple local nodes simultaneously. Each node receives and implements model updates independently, enabling parallel processing that accelerates deployment while maintaining system manageability through modular architecture.
Solution Approach 2:
The patent creates a universal model update mechanism that functions across all local nodes regardless of specific data variations. The same update package can be applied universally across the network, simplifying implementation complexity while achieving rapid system-wide deployment and consistent fraud detection capabilities.
Data Source
AI summary
A local node for updating detection models while maintaining data privacy has an aggregation module, a retraining module, an instructions module, and a sharing module. The aggregation module aggregates the data into features that describe the contents of the data. The retraining module retrains the detection model using the features by implementing an algorithm that includes at least one selected feature and a threshold for triggering an activity alert. The instructions module determines instructions for calculating the at least one selected feature from a different collection of data. The sharing module generates a package having the instructions for calculating the at least one selected feature from the different collection of data and the threshold, and transmits the package to a local node for implementation of the retrained detection model with data stored at the local node.