Neural Network Functional Subnetwork Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In complex neural networks, understanding the function of subnetworks becomes inscrutable due to the unmanageable complexity, even with known network topologies, leading to lost human oversight and unclear functionality.
Innovation Solution
The method involves defining functional edges within neural networks that specify information propagation direction, generating functional subgraphs based on structural connections and directional communication, and analyzing these subgraphs using topological methods to characterize and distinguish between inputs, thereby identifying functional subnetworks and their responses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If the complete network topology is known in complex neural networks, then structural information is available, but human oversight is lost and the function of subnetworks becomes inscrutable
Solution Approach 1:
The patent divides the complex neural network into smaller functional subnetworks based on temporal activity patterns. By segmenting the network into manageable subunits that are active during specific time periods, the system maintains comprehensibility while analyzing complex overall structures. This allows human oversight to be preserved through analysis of individual subnetworks rather than being overwhelmed by the complete network topology.
2Productivity
If neural networks have increased complexity with more vertices and connections, then processing capability is improved, but analysis of topology and weights becomes untenable
Solution Approach 1:
The patent introduces a dynamic approach by defining functional subnetworks that change over time based on vertex activity patterns. Instead of static analysis of the entire network, the system dynamically identifies active subnetworks at different time periods, making analysis tractable while preserving the processing capabilities of the full complex network.
3Measurement precision
If functional subnetworks are defined temporally with directional information flow, then functional characterization is improved, but definition and analysis complexity increases
Solution Approach 1:
The patent performs preliminary actions by first identifying the complete network topology and vertex connections before temporal analysis. This preliminary structural characterization provides a foundation for subsequent temporal analysis, reducing the overall complexity by breaking down the problem into preparatory and analysis phases.
Data Source
AI summary
In one aspect, a method includes generating a functional subgraph of a network from a structural graph of the network. The structural graph comprises a set of vertices and structural connections between the vertices. Generating the functional subgraph includes identifying a directed functional edge of the functional subgraph based on presence of structural connection and directional communication of information across the same structural connection.