Classifier Re-Learning Using Boundary-Focused Pseudo Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing classifiers generated through machine learning face challenges in maintaining accurate classification results due to re-learning, leading to potential erroneous outputs, especially when evaluated using evaluation data strictly defined for product deployment, and are inefficient for multidimensional signals like images.
Innovation Solution
A computing machine generates a classifier by analyzing the distribution of learning data in a feature space, identifying boundaries where classification results change, and creating pseudo input data near these boundaries to improve classification accuracy through re-learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If re-learning is performed to improve classification accuracy for previously misclassified samples, then classification accuracy for new samples is improved, but classification accuracy for evaluation data deteriorates
Solution Approach 1:
The patent applies local quality by generating pseudo-learning data specifically at boundary regions where classification results change, rather than uniformly across the entire feature space. This targeted approach improves classification accuracy in critical regions while preserving the original classifier's performance on evaluation data, thus resolving the contradiction between improving accuracy and maintaining reliability.
Solution Approach 2:
The patent performs preliminary action by generating pseudo-learning data in advance based on the original learning data and classifier, identifying boundary regions before re-learning. This preparatory step ensures that re-learning is focused on specific areas needing improvement, preventing deterioration of classification accuracy on evaluation data while still improving overall performance.
2Measurement precision
If a complex classifier structure is used to improve classification accuracy, then classification accuracy is improved, but understanding and debugging becomes difficult
Solution Approach 1:
The patent uses copying by creating pseudo-learning data that replicates the characteristics of real learning data at boundary regions. This allows the system to work with simplified representations of complex classification scenarios, improving accuracy without requiring increasingly complex classifier structures, thus resolving the contradiction between accuracy and complexity.
3Adaptability or versatility
If traditional re-learning methods are used to adapt to new containers and samples, then adaptability is improved, but time and resources for re-learning are consumed
Solution Approach 1:
The patent performs preliminary action by pre-generating pseudo-learning data at boundary regions before actual re-learning is needed. When adapting to new containers or samples, this pre-prepared data enables faster re-learning processes, reducing the time and computational resources required while maintaining high adaptability to new conditions.
Solution Approach 2:
The patent applies local quality by focusing re-learning efforts only on boundary regions where classification changes occur, rather than re-learning across the entire feature space. This targeted approach significantly reduces the time and resources required for adaptation to new containers and samples while maintaining high adaptability.
Data Source
AI summary
A computing machine includes: a storage unit that stores learning data; a learning unit that executes learning processing for generating a classifier by using the learning data; and a generation unit that generates the learning data, and the generation unit calculates a feature amount vector handled by the classifier by using the learning data stored in the storage unit, analyzes the distribution ox the learning data in a feature amount space on the basis of the feature amount vector to specify a boundary where the classification result of the classifier changes in the feature amount space, and generates new learning data by using the learning data existing in the vicinity of the boundary.