Vector Neural Network Feature Spectrum for Class Determination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models, such as capsule networks, often fail to accurately reflect fine features of input data in class determination results due to omission during transmission, leading to incorrect outcomes.
Innovation Solution
A method involving a vector neural network with multiple vector neuron layers, where a known feature spectrum group is prepared from training data outputs, and a feature spectrum is calculated for new data, allowing for class similarity calculation and determination based on these similarities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a vector neural network type machine learning model is used for class determination, then the processing speed and automation are improved, but the fine features of input data are omitted during transmission leading to incorrect class determination results
Solution Approach 1:
The patent applies preliminary action by calculating and storing the feature spectrum from the specific layer output during the training phase. This pre-calculated feature spectrum serves as a reference that preserves fine feature information, which is then compared with the feature spectrum of new data during class determination, preventing information loss that would occur in traditional end-to-end vector neural networks.
2Device complexity
If the class determination is performed using the output layer directly, then the device complexity is reduced, but the measurement precision of class determination deteriorates due to omitted fine features
Solution Approach 1:
The patent introduces an intermediary approach by using the feature spectrum calculated from the specific layer output as a mediator between the input data and the class determination. This feature spectrum acts as an intermediate representation that preserves fine feature information while enabling comparison with reference feature spectra, thereby improving measurement precision without significantly increasing device complexity.
3Measurement precision
If fine features of input data are preserved through intermediate layer analysis, then the class determination accuracy is improved, but the calculation complexity and processing time increase
Solution Approach 1:
The patent applies the extraction principle by taking out only the necessary feature spectrum from the specific layer output, rather than processing or analyzing all intermediate layer data. This selective extraction of the feature spectrum preserves fine feature information while minimizing calculation complexity, as it focuses computation on the most relevant features needed for accurate class determination.
Data Source
AI summary
A class determination method includes: step (a): preparing, for each of a plurality of classes, a known feature spectrum group obtained when a plurality of pieces of training data are input to a vector neural network type machine learning model; and step (b): executing, by using the machine learning model and the known feature spectrum group, a class determination processing on data to be determined. The step (b) includes step (b1), calculating a feature spectrum according to an input of the data to be determined to the machine learning model, step (b2), calculating a class similarity between the feature spectrum and the known feature spectrum group related to each of the plurality of classes, and step (b3), determining a class of the data to be determined according to the class similarity.