Dynamic Class Weights for Wide-Range Feature Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional class classification methods struggle to accurately classify input images when feature amounts within a class are distributed in a wide range, as they rely on a single weight vector that cannot effectively express multimodal distributions.
Innovation Solution
A class classification method that generates a continuously changing weight vector for each class using a trained generator, calculates scores based on these vectors, and classifies input data into one of multiple classes, utilizing an orthonormal matrix to enhance expressiveness and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single weight vector is used for each class, then the classifier structure is simple, but it cannot accurately classify input images when feature amounts are distributed in a wide range
Solution Approach 1:
The patent divides the single weight vector into multiple weight vectors (e.g., w_c1, w_c2, w_c3) representing different prototypes or centroids within the same class. This segmentation allows the classifier to capture multiple modes of feature distribution within a class, improving classification accuracy for wide-range distributed features while maintaining reasonable structural complexity through organized prototype representation.
2Adaptability or versatility
If multiple weight vectors are used to express multiple classes, then the expressiveness improves, but the difficulty to accurately perform class classification increases when feature amounts are distributed in a wide range
Solution Approach 1:
The patent introduces dynamic weighting where each weight vector is associated with a learnable weight parameter (e.g., a1, a2, a3) that can be adjusted based on the input features. This dynamic mechanism allows the system to adaptively combine multiple weight vectors according to the specific characteristics of the input data, improving both expressiveness and classification accuracy for wide-range distributed features.
Solution Approach 2:
The patent changes the parameters of the weight vectors through learning processes, allowing them to be optimized for specific classification tasks. By adjusting the weight vectors and their associated parameters during training, the system can accurately capture the distribution characteristics of features within a wide range, resolving the contradiction between expressiveness and classification accuracy.
3Productivity
If conventional classification methods are used, then the processing is simple, but the classification accuracy deteriorates when feature amounts within a class are distributed in a wide range
Solution Approach 1:
The patent performs preliminary actions during the training phase by pre-computing multiple weight vectors and their associated parameters for each class. This preliminary preparation allows the system to efficiently handle wide-range distributed features during actual classification tasks, maintaining processing efficiency while significantly improving classification accuracy through the pre-learned multi-mode representations.
Data Source
AI summary
This class classification device: acquires feature vectors extracted from input data; generates, by using a trained generator, weight vectors that continuously change according to the values of the feature vectors, for a plurality of classes to be classified; calculates, by using a trained calculator, scores for the plurality of classes on the basis of the feature vectors and a plurality of the weight vectors generated for the plurality of classes; classifies a classification target of the input data into any of the plurality of classes on the basis of a plurality of the scores calculated for the plurality of classes; and outputs the classification result.