Overlay Data Store Search for Low-Latency Transaction Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing transaction processing systems face challenges in implementing machine learning models for real-time, high-volume digital transactions due to constraints such as millisecond latencies, large-scale data processing, data siloing, and the need for high availability and security, leading to inaccurate fraud detection and high false positives.
Innovation Solution
A machine learning system that dynamically updates models for real-time transaction processing, allowing for fast inference and parallelization, and enables data storage without duplication, using separate data sets with atomic changes to reduce processing latency and increase throughput.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If traditional mainframe architectures and siloed data storage systems are used for transaction processing, then data security and system stability are maintained, but processing speed, latency, and throughput are severely limited
Solution Approach 1:
The patent segments the monolithic data storage system into multiple independent data stores, each handling specific data types (e.g., transaction data, user data, fraud data). This segmentation enables parallel processing across multiple stores, improving throughput while maintaining security through isolated access controls. Each data store can be independently optimized and scaled without affecting the entire system.
Solution Approach 2:
The patent introduces a temporal dimension to data storage by implementing versioned data stores that maintain historical states alongside current data. This allows the system to process transactions in real-time while preserving audit trails and enabling rollback capabilities, thus improving processing speed without sacrificing security or data integrity.
2Measurement precision
If machine learning models are implemented for fraud detection in real-time, then fraud detection accuracy improves, but processing latency increases beyond acceptable thresholds
Solution Approach 1:
The patent implements preliminary action by pre-computing fraud risk scores and maintaining pre-trained machine learning models in the data stores. During real-time transaction processing, the system performs fast lookups of pre-computed risk indicators rather than executing full machine learning inference, thereby maintaining high detection accuracy while reducing latency to acceptable levels.
Solution Approach 2:
The patent replaces traditional mechanical rule-based fraud detection systems with intelligent machine learning models that automatically learn fraud patterns. The system substitutes manual rule configuration with automated model training and deployment, achieving superior detection accuracy while the underlying data store architecture optimizes for fast retrieval of model predictions.
3Reliability
If data is stored in siloed partitions for security reasons, then data security is maintained, but system throughput and query performance deteriorate
Solution Approach 1:
The patent implements a universal data store architecture where each data store is designed to serve multiple functions: storing current and historical data, supporting multiple query patterns, enabling parallel access, and providing security through configurable access controls. This multi-functionality allows the system to maintain security partitions while achieving high throughput through efficient resource utilization.
Solution Approach 2:
The patent introduces data store metadata and indexing structures as intermediaries that enable efficient querying across partitioned data. These intermediaries maintain security boundaries while providing optimized access paths that improve throughput, allowing the system to simultaneously achieve both security isolation and high-performance data retrieval.
4Measurement precision
If traditional rule-based fraud detection systems are used, then system simplicity is maintained, but detection accuracy and false positive rates are unacceptable
Solution Approach 1:
The patent implements self-service by enabling machine learning models to automatically learn fraud patterns from historical data and continuously improve detection accuracy without manual intervention. The system autonomously trains models, optimizes parameters, and adapts to emerging fraud techniques, achieving high detection accuracy while reducing the operational complexity of manual rule maintenance.
Solution Approach 2:
The patent leverages parameter changes by using machine learning models that dynamically adjust detection thresholds and weighting parameters based on learned patterns. This allows the system to achieve high detection accuracy by adapting parameters to current fraud trends rather than relying on static rules, while the underlying data store manages the complexity of parameter storage and retrieval.
Data Source
AI summary
A data item is searched for in first and/or second data sets of a data store. If it is found in the first data set and if it was updated in the first data set after the second data set became an overlay of the first data set, first data stored in association with the data item in the first data set is returned. If it is found in the first and second data sets and if the second data set became an overlay of the first data set after the data item was updated in the first data set, second data stored in association with the data item in the second data set is returned. The second data set is identified based on overlay metadata, indicative of the second data set being an overlay of the first data set.