Dynamic Training Batch Adjustment for Classification Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Classification models face challenges in balancing training data, leading to inconsistent learning across classes, resulting in suboptimal accuracy and efficiency during the training process.
Innovation Solution
The solution involves dynamically adjusting the number of samples in training batches based on the accuracy of each class, increasing samples for underperforming classes and decreasing samples for well-performing classes, using techniques such as proportional analysis and accuracy thresholds to ensure all classes reach a desired accuracy level.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the number of training samples for each class is increased to improve classification accuracy, then the training time and computational resources increase
Solution Approach 1:
The patent applies local quality by differentiating the treatment of training samples based on class-specific accuracy metrics. Instead of uniformly increasing samples for all classes, the system identifies underperforming classes and selectively increases their sample representation while maintaining or reducing samples for well-performing classes. This targeted approach improves overall classification accuracy without proportionally increasing total training time across all classes.
Solution Approach 2:
The patent implements dynamics by making the training sample allocation adaptive and changeable throughout the training process. The system continuously monitors classification accuracy for each class and dynamically adjusts the number of training samples accordingly. This dynamic adjustment allows the training process to respond to real-time performance feedback, optimizing the balance between accuracy improvement and training time investment.
2Productivity
If the training data is balanced across all classes, then the training process becomes more efficient, but classes with initially lower accuracy may still underperform
Solution Approach 1:
The patent applies feedback by continuously monitoring the classification accuracy of each class during training and using this information to adjust subsequent training sample allocation. The system measures performance metrics for each class, compares them against target thresholds, and feeds this information back into the sample selection process. This closed-loop feedback mechanism ensures that underperforming classes receive additional training attention while maintaining overall training efficiency.
Solution Approach 2:
The patent implements parameter changes by modifying the weight or quantity of training samples assigned to different classes based on their performance parameters. The system adjusts sample allocation parameters dynamically, increasing the proportion of samples from underperforming classes while reducing or maintaining samples from well-performing classes. This parameter adjustment strategy rebalances training efficiency with targeted accuracy improvement for struggling classes.
Data Source
AI summary
Various embodiments are generally directed to techniques of adjusting the combination of the samples in a training batch or training set. Embodiments include techniques to determine an accuracy for each class of a classification model, for example. Based on the determined accuracies, the combination of the samples in the training batch may be adjusted or modified to improve the training of the classification model.