Neural Network Self-Diagnosis via Sensitive Node Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural networks cannot detect abnormalities within themselves and often become redundant in scale when attempting to do so, leading to inefficiencies in resources and performance.
Innovation Solution
A neural network optimization system that includes a definition data analysis unit, an internode dependence degree analysis unit, and a sensitive node extraction unit to identify and add diagnosis circuits only to sensitive nodes, allowing for self-diagnosis while minimizing redundancy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a neural network is designed to detect its own abnormalities, then the reliability of the system is improved, but the device complexity and scale increase significantly
Solution Approach 1:
The patent segments the neural network into multiple layers and identifies specific sensitive nodes within each layer that have high impact on output accuracy. Instead of monitoring all nodes, the system divides the network into manageable segments and focuses diagnostic resources on critical segments, thereby reducing overall complexity while maintaining detection capability.
Solution Approach 2:
The patent applies local quality by assigning different monitoring priorities to different nodes based on their sensitivity and impact on output. Critical nodes with high dependence degrees receive focused monitoring resources, while less critical nodes use standard monitoring. This differentiated approach reduces the overall scale of the monitoring system while maintaining effective abnormality detection.
2Measurement precision
If comprehensive monitoring of all neural network nodes is implemented, then the measurement precision of abnormalities is improved, but the loss of energy and calculation resources increases
Solution Approach 1:
The patent changes the monitoring parameters by calculating dependence degrees and sensitivity metrics for each node, then using these parameters to prioritize monitoring efforts. Nodes with higher dependence degrees and sensitivity values are monitored with greater precision and frequency, while nodes with lower values receive reduced monitoring. This parameter-based prioritization maintains detection accuracy for critical abnormalities while reducing overall energy consumption.
3Adaptability or versatility
If the neural network structure is expanded to include self-diagnosis functionality, then the versatility of the system is improved, but the manufacturing precision and resource efficiency deteriorate
Solution Approach 1:
The patent performs preliminary action by pre-calculating and storing dependence degree information and sensitivity metrics for each node during the network configuration phase. This preliminary analysis creates a roadmap for efficient monitoring that guides subsequent real-time diagnosis operations. By preparing this information in advance, the system avoids complex real-time calculations and simplifies the implementation of self-diagnosis functionality.
Data Source
AI summary
A neural network that can detect abnormality of itself while suppressing redundancy of a scale is realized. A neural network optimization system includes a definition data analysis unit configured to analyze learned neural network definition data, an internode dependence degree analysis unit configured to generate dependence degree information indicating an internode dependence degree in a learned neural network defined by the learned neural network definition data, based on an analysis result of the learned neural network definition data, and a sensitive node extraction unit configured to extract a sensitive node in the learned neural network based on the dependence degree information.