Neural Network Cluster Analysis via Squared Error Metrics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for simplifying multilayer neural networks fail to provide a quantitative understanding of the role of each cluster in the network, leading to unstable results due to threshold dependencies and lack of clarity on data processing roles.
Innovation Solution
An analyzing device and method that calculates the sum of squared errors between unit outputs and input/output data dimensions, allowing for a quantitative analysis of cluster contributions and relearning the neural network with noise adjustments based on these calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If threshold processing is used to simplify the neural network structure by expressing multiple connections as one connection, then the structure simplification is achieved, but the examination results become unstable and change greatly depending on threshold setting
Solution Approach 1:
The patent changes the parameter used for connection representation from binary threshold-based connections to continuous connection strengths. Instead of using a fixed threshold to determine whether a connection exists, the system uses the actual connection strength values (weights) to represent connections, allowing for a more nuanced and stable representation that doesn't abruptly change with small threshold variations.
2Device complexity
If existing techniques are used to extract cluster structure, then structure simplification is achieved, but quantitative understanding of each cluster's role is not obtained
Solution Approach 1:
The patent introduces connection strength as an intermediary element that bridges the gap between structural simplification and quantitative analysis. By using connection strengths to represent relationships between clusters, the system maintains both the simplified cluster structure and the quantitative information needed to analyze cluster roles and contributions.
3Ease of manufacture
If threshold processing is applied to simplify connections, then computational ease is improved, but manufacturing precision of the analysis results deteriorates
Solution Approach 1:
Instead of applying full threshold processing that would simplify computation but lose precision, the patent uses a partial approach by incorporating connection strength information without complete thresholding. This allows the system to maintain high analysis precision while still benefiting from some level of computational simplification through the cluster-based representation.
Data Source
AI summary
The present invention enables the structure of a neural network to be quantitatively analyzed. An analyzing unit calculates, for each of combinations of a dimension of input data and a cluster, a sum of squared errors between an output of each unit belonging to the cluster when a value of the dimension of the input data is replaced with an average value of the dimension of the input data included in learning data and an output of each unit belonging to the cluster for the input data before replacement as a relationship between the combinations, and calculates, for each of combinations of the cluster and a dimension of output data, a squared error between the value of the dimension of the output data when an output value of each unit belonging to the cluster is replaced with an average output value of each unit of the cluster when the input data included in the learning data was input and the value of the dimension of the output data before replacement as a relationship between the combinations.