Verification Neural Network for Reliable Classification Outputs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural networks face challenges in determining the reliability of classification results when input data differs from training data, necessitating additional processes to verify the accuracy of the classification outcomes.
Innovation Solution
A verification neural network is employed to determine the reliability of classification results by using intermediate output values from a classification neural network, trained on attribute information and distance between training and sample data, with reliability models adjusting based on similarity and gradient directions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a classification neural network is used to classify input data, then classification speed and productivity are improved, but the reliability of classification results deteriorates when input data differs from training data
Solution Approach 1:
A verification neural network is introduced as an intermediary component between the classification neural network and the final output. The verification network receives intermediate output values from the classification network and determines whether to trust the classification result, thereby resolving the contradiction between speed and reliability by adding a specialized verification module rather than slowing down the entire classification process
Solution Approach 2:
The system is divided into two independent neural networks: a classification neural network for fast classification and a verification neural network for reliability assessment. This segmentation allows each network to specialize in its function, maintaining high productivity in classification while ensuring reliability through dedicated verification
2Reliability
If a verification process is added to determine reliability of classification results, then reliability is improved, but device complexity and processing time increase
Solution Approach 1:
The verification neural network is trained in advance on training data to learn the characteristics of reliable classifications. During inference, it quickly evaluates intermediate output values without requiring complex real-time analysis, thereby improving reliability while minimizing additional processing time and system complexity
3Measurement precision
If intermediate output values from hidden layers are used for verification, then measurement precision of reliability assessment is improved, but device complexity increases
Solution Approach 1:
The verification neural network is designed to process only specific intermediate output values from particular hidden layers of the classification network, rather than analyzing all possible features. This localized approach focuses verification resources on the most informative intermediate representations, improving measurement precision while controlling complexity
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A classification result verification method and a classification result learning method which use a verification neural network, and a computing device for performing the methods are disclosed. The classification result verification method inputs, to a verification neural network, an intermediate output value for input data output from a classification neural network, and allows the verification neural network to output a reliability of a classification result for the input data of the classification neural network using the intermediate output value, to verify the classification result for the input data.