Learning Apparatus for Classification Metric Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing classifier learning methods require manual attribution of learning samples, increasing costs and limiting the number of classification metrics that can be learned, as they rely on human-defined attributes like gender or race.
Innovation Solution
A learning apparatus that automatically selects groups of learning samples with varying categories and learns classification metrics to generate an evaluation metric, reducing the need for manual attribution and enabling the learning of multiple classification metrics without upper limits.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual attribution of attributes is performed for each learning sample, then classification metrics can be learned, but costs increase and the number of learnable classification metrics is limited
Solution Approach 1:
The system automatically selects groups of learning samples and learns classification metrics without requiring manual attribution of attributes to each sample. The learning sample selection unit autonomously performs sample selection based on category information, and the classification metric learning unit automatically learns metrics from these selected groups, eliminating the need for human annotators to assign attributes to individual samples.
Solution Approach 2:
The system can learn multiple different types of classification metrics simultaneously without being limited by manual attribution constraints. By using unattributed learning samples with only category information and automatically learning multiple classification metrics from selected sample groups, the system achieves multi-functionality in terms of the number and variety of classification metrics that can be learned.
2Reliability
If multiple classification metrics are learned using manual attribution, then recognition accuracy improves, but the cost and time required increase significantly
Solution Approach 1:
The system autonomously performs the entire workflow from learning sample selection to classification metric learning without human intervention. The learning sample selection unit automatically selects appropriate groups of samples, and the classification metric learning unit learns multiple metrics from these groups, completely eliminating the time-consuming manual attribution process while maintaining recognition accuracy.
Solution Approach 2:
The system prepares learning samples by category before the learning process begins, storing them in advance in the learning sample storage unit. This preliminary organization by category enables the automatic selection and learning processes to proceed efficiently without requiring manual attribution during the actual learning phase, thereby reducing time loss.
3Adaptability or versatility
If attributes are manually assigned to learning samples, then two-class classifiers can be learned, but the system complexity and operational burden increase
Solution Approach 1:
The system automatically performs sample selection and classification metric learning without requiring operators to manually assign attributes to each learning sample. The learning sample selection unit autonomously selects groups of samples based on category information, and the classification metric learning unit automatically learns metrics from these groups, significantly simplifying operation while maintaining classification versatility.
Solution Approach 2:
The system extracts only the necessary category information from learning samples and stores it in the learning sample storage unit, removing the need for detailed attribute annotations. By using only category labels instead of full attribute descriptions, the system reduces operational complexity while preserving the ability to learn multiple classification metrics.
Data Source
AI summary
According to an embodiment, a learning apparatus includes a learning sample storage unit configured to store therein a plurality of learning samples that are classified into a plurality of categories; a selecting unit configured to perform, plural times, a selection process of selecting a plurality of groups each including one or more learning samples from the learning sample storage unit; and a learning unit configured to learn a classification metric for classifying the plurality of groups selected in each selection process and generate an evaluation metric including the learned classification metrics.