Classifier Array Individuality for Pattern Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Classifiers in pattern recognition often misclassify objects, leading to a need for improved classification accuracy.
Innovation Solution
Designing classifiers with individuality by randomly selecting subsets of feature vectors and training them with weighted inputs to correct each other's mistakes, resulting in a combined classifier with unique behavior that enhances overall accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single classifier is trained to classify objects into predefined classes, then the classification process is simple and efficient, but the classifier misclassifies objects some of the time
Solution Approach 1:
The patent divides a single classifier into multiple individual classifiers, each trained on different random subsets of features. This segmentation allows each classifier to have specialized knowledge while the collective system achieves higher overall accuracy through complementary strengths and error cancellation.
Solution Approach 2:
The patent merges multiple individual classifiers into a unified system where each classifier contributes to the overall classification task. The combined system leverages the diverse feature subsets learned by each individual classifier to achieve superior classification performance compared to any single classifier alone.
2Reliability
If multiple classifiers are used to improve classification accuracy, then misclassifications can be corrected, but the complexity of the system increases
Solution Approach 1:
The system segments the classification task among multiple independent classifiers, each handling specific feature subsets. This segmentation enables parallel processing and error distribution, where misclassifications by one classifier can be corrected by others, improving overall reliability without requiring complex coordination mechanisms.
Solution Approach 2:
The patent introduces randomness as a key parameter in classifier generation, where each classifier is trained on different random subsets of features. This parameter change creates diversity among classifiers while maintaining simplicity in individual classifier structure, achieving improved accuracy through statistical diversity rather than structural complexity.
Data Source
AI summary
The systems and methods of the invention described herein improve classification accuracy by producing classifiers with individuality, in which each classifier exhibits its own distinctive behavior. A method according to one exemplary embodiment produces each classifier with individuality by randomly selecting subsets of a feature vector and using the randomly selected subsets of the feature vector in the design of the classifier. Because different subsets of the feature vector are used for each classifier, each classifier exhibits its own distinctive behavior or individuality. The classifiers with individuality improve classification accuracy, for example, when used as classifiers in a classifier array. This is because the individuality among the classifiers in the array ensures that a misclassification committed by one of the classifiers will not be repeated by other classifiers in the array, thereby improving the overall accuracy of the classifier array.