Neural Network Classifier Verification via Interface Region Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural networks face challenges in accurately classifying data sets with overlapping categories due to significant overlap between data distributions, making proper classification difficult and complicating the certification process for AI/ML agents, particularly in aerospace applications.
Innovation Solution
An automated system evaluates the efficiency of a neural network classifier by identifying classification outputs that fall between identifiable classification parameter groupings, using support vectors to define a separation plane and determine the percentage of data points in an interface region, providing a verified classifier if this percentage is below a threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If neural networks classify data by quantifying features, then classification capability is provided, but accuracy deteriorates when data categories overlap significantly
Solution Approach 1:
The patent replaces traditional mechanical classification approaches with support vector machines that use mathematical hyperplanes and margin maximization to achieve better separation of overlapping data categories, improving accuracy while maintaining classification capability
Solution Approach 2:
The patent changes the classification parameters by introducing margin-based evaluation and interface region analysis, transforming the classification problem into a geometric optimization problem that better handles overlapping data distributions
2Productivity
If traditional AI/ML training methods are used, then model development is achieved, but verification and certification become difficult and expensive
Solution Approach 1:
The patent performs verification actions during the training process itself by continuously monitoring interface region percentages and support vector configurations, so that certification requirements are met before deployment rather than requiring separate expensive verification phases
Solution Approach 2:
The patent introduces feedback mechanisms that monitor classification performance metrics during training and provide real-time information about interface region percentages, enabling iterative improvement and automated verification of model certification requirements
3Measurement precision
If data points are classified into distinct categories, then classification clarity is improved, but misclassification occurs in interface regions between categories
Solution Approach 1:
The patent introduces support vectors as intermediary elements that define the boundary between categories and create a margin region, acting as a buffer that reduces the impact of misclassification in interface regions while maintaining clear category separation
Solution Approach 2:
The patent creates a cushioning effect by maximizing the margin between support vectors of different classes, which beforehand prevents data points from falling into ambiguous interface regions and reduces misclassification probability before it occurs
Data Source
AI summary
A method for defining neural network classifier performance includes causing processing of data by a plurality of convolutional layers of an artificial intelligence (AI) agent, where the processing of the data is caused by providing data to the AI agent, receiving one or more outputs from a first convolutional layer of the plurality of convolutional layers, generating one or more classification data points by processing the one or more outputs through a classifier of the NN that classifies the one or more outputs into two or more classifications, determining an interface region between the two or more classifications, determining a percentage of classification data points that fall into the interface region, and providing the AI agent as an AI agent with a verified classifier in response to the percentage of classification data points that fall into the interface region being below a threshold.