Adaptive Transaction Model Orchestration for Private Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing transaction processing systems struggle with inefficiencies in identifying potentially anomalous activities due to reliance on static rules and manual processing, which can lead to false positives, missed detections, and inability to adapt to changing transaction patterns.
Innovation Solution
A system that orchestrates iterative updates to machine learning models deployed across multiple end-user devices, using model parameters and weights to refine transaction models without exchanging transaction data, and applies peer group classifications for dynamic and contextualized anomaly detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If static rules and manual processing are used for transaction monitoring, then system complexity is reduced, but detection accuracy and adaptability to changing transaction patterns deteriorate
Solution Approach 1:
The patent implements dynamic transaction monitoring by deploying machine learning models that automatically adapt to changing transaction patterns. The system continuously learns from new data and adjusts detection criteria without requiring manual rule updates, enabling the system to evolve with emerging transaction behaviors while maintaining operational simplicity.
Solution Approach 2:
The system incorporates feedback mechanisms where transaction monitoring outcomes are fed back into the machine learning models to continuously improve detection accuracy. The models learn from both true positives and false positives, adjusting their parameters to enhance future detection performance while reducing manual intervention requirements.
2Measurement precision
If machine learning models are deployed across multiple end-user devices with iterative updates, then detection accuracy and adaptability improve, but data security risks increase
Solution Approach 1:
The patent extracts only the necessary model parameters and weights from the machine learning models for deployment to end-user devices, while keeping the core transaction data and sensitive information centralized. This extraction approach enables accurate local detection without exposing sensitive data, as devices receive only the computational logic needed for anomaly detection.
Solution Approach 2:
The system introduces an intermediary orchestration layer that manages model updates and coordinates between centralized data storage and distributed end-user devices. This intermediary ensures that model parameters are securely transmitted and updated without exposing underlying transaction data, maintaining data security while enabling iterative model improvements across the network.
3Reliability
If manual processing rules are used, then false positives are reduced through human judgment, but processing throughput and productivity deteriorate
Solution Approach 1:
The machine learning models are designed to autonomously detect and classify suspicious transactions without requiring manual review for every case. The models self-adjust their detection thresholds and parameters based on learned patterns, automatically filtering out normal variations in transaction behavior while flagging genuine anomalies for further investigation, thereby maintaining high throughput with reduced false positives.
Solution Approach 2:
The system dynamically changes detection parameters based on learned transaction patterns and contextual information. Instead of using fixed manual rules, the models adjust sensitivity thresholds, time windows, and pattern matching criteria in real-time based on the specific transaction context, improving detection accuracy while maintaining processing speed.
4Measurement precision
If transaction data is exchanged between end-user devices for model refinement, then model performance improves, but data privacy and security are compromised
Solution Approach 1:
The patent extracts only the essential model parameters and performance metrics needed for model refinement, while deliberately excluding sensitive transaction data from exchange between devices. End-user devices share only aggregated statistics and parameter updates that improve model performance without revealing underlying customer information or transaction details.
Solution Approach 2:
Instead of exchanging original transaction data, the system uses copies in the form of anonymized aggregates and synthesized patterns that capture the essential learning signals. These synthetic representations allow models to learn from collective experience across multiple devices without exposing actual customer data, maintaining privacy while improving performance.
Data Source
AI summary
Systems and techniques are described for orchestrating iterative updates to machine learning models (e.g., transaction models) deployed to multiple end-user devices. In some implementations, output data generated by a first transaction model deployed at a first end-user device is obtained. The first transaction model is trained to apply a set of evidence factors to identify potentially anomalous activity associated with a first target entity. An adjustment for a second transaction model deployed at a second end-user device is determined. The second transaction model is trained to apply the set of evidence factors to identify potentially anomalous activity associated with a second target entity determined to be similar to the first target entity. A model update for the second transaction model is generated. The model update specifies a change to the second transaction model. The model update is provided for output to the second end-user device.