Graph Convolution Anomaly Detection for Spurious Data Patterns
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems lack robust methods for detecting spurious data patterns in real-time, particularly in the context of financial transactions, which can lead to fraudulent activities.
Innovation Solution
A method and apparatus using a graph network methodology and neural network topology for detecting anomalous data patterns, combined with data preprocessing techniques to reduce dimensionality and enhance detection sensitivity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional detection methods are used, then the system is simple to implement, but the detection reliability is insufficient for robustly detecting spurious data patterns
Solution Approach 1:
The patent introduces graph network representations as an intermediary structure between raw data and neural network detectors. Data points are converted into graphs with nodes and edges that capture relationships, serving as a mediator that enhances the detector's ability to identify spurious patterns without directly increasing the neural network's structural complexity
Solution Approach 2:
The patent transforms data from traditional tabular or vector representations into graph-structured representations, adding a relational dimension. This dimensional change enables the system to capture complex dependencies and patterns that are not visible in conventional data formats, improving detection reliability
2Productivity
If data dimensionality is reduced through preprocessing, then the neural network processing speed increases, but information loss may occur
Solution Approach 1:
The patent extracts and removes only the most relevant features and relationships from the data during graph construction, filtering out redundant information while preserving critical patterns. This selective extraction reduces dimensionality for faster processing without sacrificing essential information needed for detection
Solution Approach 2:
The patent changes the representation parameters of data from raw values to graph-structured formats with specific node and edge feature parameters. This parameter transformation enables efficient processing by the neural network while maintaining information integrity through careful feature selection and relationship encoding
3Measurement precision
If graph network methodology is applied to detect outliers, then the detection precision improves, but the computational time increases
Solution Approach 1:
The patent performs preliminary graph construction and feature extraction during an offline phase, pre-processing data into graph representations before actual detection occurs. This preliminary action reduces the computational burden during real-time detection, improving precision without significantly increasing online computational time
Solution Approach 2:
The patent segments the detection process into distinct stages: graph construction, feature extraction, and neural network classification. This segmentation allows each component to be optimized independently, with graph operations performed in advance and only essential computations performed during real-time detection
Data Source
AI summary
Disclosed are implementations that include a method for detecting anomalous data, including converting a set of data values representative of a multi-dimensional item into a nodes-and-edges graph representation of the item, applying a graph convolution process to the graph representation to generate a transformed graph representation for the item comprising a resultant transformed configuration of the nodes and edges representing the item, and determining, based on the transformed configuration, a probability that the item is anomalous. Another example method includes receiving input data at a neural network circuit comprising a plurality of node layers, with each of the plurality of node layers comprising respective one or more nodes, with the neural network circuit further comprising adjustable weighted connections connecting at least some nodes in different layers of the plurality of node layers. The method further includes removing one or more of the weighted connections at one or more time instances.