Shared Detection Model Updates With Privacy-Preserving Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing analytical models struggle to update rapidly and securely across multiple local nodes while maintaining data privacy, especially in the face of evolving threats like banking fraud, where third parties attempt to evade detection.
Innovation Solution
A system and method for updating detection models that allows for rapid distribution and implementation across local nodes, using a local node with modules for data collection, retraining, and feature calculation, ensuring data privacy through indeterminable general features and values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If data is shared across multiple local nodes for model training, then model accuracy and detection capability are improved, but data privacy and security are compromised
Solution Approach 1:
The patent extracts only the necessary feature information from raw data while leaving the sensitive raw data localized at each node. General features are calculated and shared across the network, enabling collaborative model training without exposing proprietary or sensitive raw data, thus maintaining both model accuracy and data privacy
Solution Approach 2:
The patent segments the data processing workflow into local feature extraction at each node and centralized model aggregation. Each node independently calculates general features from its local data, then these features are shared and aggregated to train the global model, dividing the sensitive data handling into isolated, privacy-preserving segments
2Productivity
If detection models are updated rapidly across multiple local nodes, then responsiveness to evolving threats is improved, but system complexity and coordination overhead increase
Solution Approach 1:
The patent implements preliminary action by pre-calculating and sharing general feature extraction instructions with all local nodes before model updates are needed. Each node prepares its feature calculation capabilities in advance, so when model updates are required, the nodes can rapidly participate in training without coordination delays for data preparation
Solution Approach 2:
The patent creates a universal feature extraction framework that works across all local nodes regardless of their specific data types or domains. The same general feature calculation instructions can be applied universally at each node, simplifying coordination and enabling rapid model updates across diverse distributed systems
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A local node for updating detection models while maintaining data privacy has a sharing module configured to receive instructions for calculating at least one general feature from data stored at the first local node, a retraining module configured to retrain the detection model to include a detection component that uses the instructions, a data collection module configured to collect data comprising customer data and transaction data stored at the first local node, and a performance module. The performance module is configured to determine a value for the at least one general feature from the collected data using the instructions, and trigger a suspicious activity alert based on the determined value and the instructions. The customer data and transaction data are indeterminable from the at least one general feature and the determined value.