Classifier Ensemble Self-Labeling for Noisy Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The processing of training data for classifiers, such as those used for object recognition in vehicle surroundings, is hindered by the need for extensive human-labeled data, which is costly and time-consuming. Additionally, existing methods may produce noisy or incorrect learning output variable values, leading to compromised classification accuracy.
Innovation Solution
A method is introduced where learning input variable values from existing learning data sets are processed multiple times by different classifiers or modifications of a classifier, allowing for the assessment of uncertainty in output variable values. If the uncertainty exceeds a predefined threshold, updated learning output variable values are obtained by resetting the classifiers to an earlier training level, thereby improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If learning data sets are labeled using human work, then the accuracy of learning output variable values is improved, but the cost and time consumption increase
Solution Approach 1:
The patent uses multiple classifiers to generate multiple copies of output variable values for the same learning input variable value. These multiple predictions are then aggregated to determine the final learning output variable value, replacing the need for expensive human labeling while maintaining accuracy through ensemble methods
Solution Approach 2:
The system performs self-labeling by using the classifiers themselves to generate learning output variable values from learning input variable values. The multiple classifiers process the same input and their aggregated results serve as the labeled output, making the system self-sufficient without requiring external human annotators
2Measurement precision
If learning data sets are labeled using human work, then the accuracy of learning output variable values is improved, but the cost increases
Solution Approach 1:
Multiple classifiers generate multiple predictions (copies) for each learning input variable value. These copies are aggregated through voting or averaging mechanisms to produce the final learning output variable value, eliminating the need for costly human labeling while maintaining high accuracy through ensemble diversity
Solution Approach 2:
The classifier system performs self-labeling by generating its own learning output variable values from learning input variable values. This self-service approach replaces expensive human annotators with automated classifier ensembles, significantly reducing costs while maintaining labeling quality
3Measurement precision
If classifiers are trained to higher epochs, then the classification accuracy is improved, but the sensitivity to noisy labels increases
Solution Approach 1:
The patent determines the epoch at which classification accuracy is maximized before proceeding with training. This preliminary identification of the optimal training point prevents over-training that would increase sensitivity to noisy labels, allowing the system to stop training at the point of peak performance rather than continuing to higher epochs
Solution Approach 2:
The system continuously monitors classification accuracy during training and uses this feedback to determine when to stop training. By detecting the point of maximum accuracy and halting further training, the system avoids the diminishing returns and increased noise sensitivity that occur at higher epochs, maintaining optimal reliability
4Measurement precision
If multiple classifiers are used to process learning input variable values, then the accuracy of learning output variable values is improved, but the device complexity increases
Solution Approach 1:
The patent combines multiple classifier predictions into a single aggregated output through voting or averaging mechanisms. This merging approach maintains the accuracy benefits of ensemble methods while simplifying the overall system architecture by consolidating multiple classifier outputs into one unified learning output variable value
Solution Approach 2:
The multiple classifiers in the system serve the universal function of generating learning output variable values from learning input variable values. Rather than requiring different specialized components, the same classifier architecture is replicated and applied universally across all learning data, reducing system complexity through standardized multi-functionality
Data Source
AI summary
A method for processing of learning data sets for a classifier. The method includes: processing learning input variable values of at least one learning data set multiple times in a non-congruent manner by one or multiple classifier(s) trained up to an epoch E2 so that they are mapped to different output variable values; ascertaining a measure for the uncertainty of these output variable values from the deviations of these output variable values; in response to the uncertainty meeting a predefined criterion, ascertaining at least one updated learning output variable value for the learning data set from one or multiple further output variable value(s) to which the classifier or the classifiers map(s) the learning input variable values after a reset to an earlier training level with epoch E1<E2.