Multi-Range Graph Anomaly Detection for Circuit Design
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional graph convolutional networks (GCNs) struggle to effectively handle various peripheral structures in graph anomaly detection, as they either fail to convolve wide-range structures in shallow networks or bury small-range structures in deep networks, limiting their ability to detect design errors in complex circuit designs.
Innovation Solution
An information processing apparatus that extracts neighboring nodes in multiple ranges for each node in a graph structure, using GCNs with varying numbers of layers to calculate anomaly scores, and determines anomaly nodes based on the maximum score across different extraction ranges, enabling comprehensive anomaly detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a shallow layered GCN is used, then the network complexity is low, but peripheral structures spanning a wide range cannot be convolved
Solution Approach 1:
The patent segments the GCN into multiple independent modules, each with a different number of layers (e.g., 1-layer, 2-layer, 3-layer GCNs). Each module processes the graph data independently to extract anomaly scores, allowing the system to capture peripheral structures at multiple scales without requiring a single complex deep network.
2Reliability
If a deep layered GCN is used, then peripheral structures in wide range can be convolved, but small-range peripheral structures are buried
Solution Approach 1:
By dividing the detection task into multiple specialized modules (shallow, medium, and deep GCNs), the patent ensures that small-range structures are captured by shallow modules while wide-range structures are captured by deep modules, preventing information loss that would occur in a single deep network.
Solution Approach 2:
The patent adds a dimensional aspect by processing the same graph data through multiple GCN modules with different depths simultaneously. This multi-dimensional processing approach allows the system to preserve both local (small-range) and global (wide-range) structural information that would otherwise be lost in a single-layered approach.
3Device complexity
If one network is used for all nodes, then the device complexity is low, but various peripheral structures cannot be handled
Solution Approach 1:
Each GCN module is designed with universal functionality to process the same type of graph data, but with different layer configurations. This allows the system to handle various peripheral structures through multiple instances of the same basic module, achieving versatility without requiring entirely different network architectures for each case.
Data Source
AI summary
An information processing apparatus includes a processor including hardware. The processor extracts neighboring nodes in two or more different extraction ranges for each node constituting input data of a graph structure. The processor calculates an anomaly score representing a degree of anomaly of the node for each extraction range based on a representation of a combination of the node and the neighboring nodes. The processor records each calculated anomaly score in a storage. The processor selects a maximum anomaly score among the anomaly scores recorded in the storage. The processor determines an anomaly node in the input data of the graph structure based on the selected maximum anomaly score. The processor outputs information of the anomaly node.