Ensemble Model Weighting for Data Labeling Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neural network models face challenges in accurately inferring classes for data labeling, particularly in semiconductor manufacturing processes, due to variations in model confidence and consistency, which affect the reliability of classification results.
Innovation Solution
An ensemble model is developed that determines weight data by integrating model confidence and consistency data for each neural network model, allowing for the generation of classification result data and score data that reflects the relative contributions of each model to overall inference, thereby improving classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single neural network model is used for classification, then the device complexity is low, but the classification accuracy and reliability are insufficient due to variations in model confidence and consistency
Solution Approach 1:
The patent combines multiple neural network models into an ensemble model where each model processes the same input data and their predictions are aggregated. The ensemble model integrates the strengths of individual models to achieve higher classification accuracy and reliability, resolving the contradiction between model complexity and performance by distributing the classification task across multiple specialized models.
Solution Approach 2:
The patent introduces weight parameters that dynamically adjust the contribution of each neural network model based on its confidence and consistency metrics. By changing the parameter values (weights) according to model performance, the system optimizes the ensemble output to maximize classification accuracy while maintaining manageable complexity through selective weighting.
2Reliability
If multiple neural network models are combined in an ensemble, then the classification accuracy improves, but the computational resources and processing time increase
Solution Approach 1:
The patent dynamically adjusts weight parameters for each model based on confidence and consistency metrics, allowing the system to allocate computational resources more efficiently. Models with higher confidence and consistency receive greater weight, maximizing classification accuracy while minimizing unnecessary computational expenditure on less reliable models.
Solution Approach 2:
The patent applies different weighting strategies to different models within the ensemble based on their individual performance characteristics. Each model's contribution is locally optimized according to its specific strengths and weaknesses, allowing the system to achieve high overall accuracy while efficiently managing computational resources by emphasizing the most reliable models for each classification task.
3Measurement precision
If weights are assigned to individual classes for each model, then the classification precision improves, but the device complexity increases due to the weight matrix management
Solution Approach 1:
The patent introduces class-specific weight parameters that adjust the contribution of each model's prediction for each class independently. This allows fine-grained control over classification precision by optimizing weights for each model-class combination, while the systematic approach to weight management through confidence and consistency metrics keeps the complexity manageable.
Solution Approach 2:
The patent employs feedback mechanisms where the weight parameters are continuously adjusted based on model confidence and consistency measurements. This feedback loop allows the system to automatically optimize classification precision by learning which models and classes should be weighted more heavily, reducing the need for manual weight management while improving precision through data-driven adjustments.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method and device for labeling are provided. A labeling method includes: determining inference performance features of respective neural network models included in an ensemble model, wherein the inference performance features correspond to performance of the neural network models with respect to inferring classes of the ensemble model; based on the inference performance features, determining weights for each of the classes for each of the neural network models, wherein the weights are not weights of nodes of the neural network models; generating classification result data by performing a classification inference operation on labeling target inputs by the neural network models; determining score data representing confidences for each of the classes for the labeling target inputs by applying weights of the weight data to the classification result data; and measuring classification accuracy of the classification operation for the labeling target inputs based on the score data.