AI Classifier Training Using Unsupervised Clustering to Reduce Bias
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AI classifiers for electromagnetic sensor detection systems can introduce unintended bias due to human selection of training images, leading to suboptimal performance in detecting concealed objects of interest, such as weapons or explosives, in security screening processes.
Innovation Solution
The development of a machine learning-based AI classifier training method that utilizes a combination of raw electromagnetic scan data and visible appearance improvement processes to generate a training set, including supervised and unsupervised training techniques, to reduce bias and enhance detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If human-selected training images are used to train AI classifiers, then the classifier can be trained with available data, but unintended bias is introduced leading to suboptimal detection performance
Solution Approach 1:
The system uses unsupervised learning algorithms to automatically identify and select representative training samples from raw electromagnetic scan data without human intervention. The algorithm autonomously clusters data, identifies outliers, and selects training images that represent the full diversity of the data distribution, eliminating human bias while maintaining training effectiveness
Solution Approach 2:
The manual human process of selecting training images is replaced with automated machine learning algorithms including unsupervised clustering and outlier detection. These computational methods objectively identify representative samples based on data characteristics rather than human perception, substituting the biased mechanical selection process with an unbiased automated system
2Ease of manufacture
If manual selection of classification parameters is performed, then the classification process can be customized, but significant time and effort are required
Solution Approach 1:
The system performs automated parameter optimization where the machine learning algorithm independently selects and tunes classification parameters based on the training data characteristics. The system self-configures feature extraction parameters, clustering parameters, and classification thresholds without requiring manual intervention, thereby eliminating the time-consuming manual parameter selection process while maintaining optimal classification performance
3Measurement precision
If conventional classifiers with manual parameter selection are used, then the system structure is simpler, but the detection accuracy is reduced
Solution Approach 1:
Conventional manual classification methods are replaced with automated machine learning systems that use unsupervised clustering, outlier detection, and automated parameter optimization. These computational mechanisms automatically identify patterns and select parameters, replacing simple but inaccurate manual methods with complex but accurate automated systems, achieving superior detection accuracy despite increased computational complexity
Data Source
AI summary
Systems and methods for training an artificial intelligence (AI) classifier of scanned items. The items may include a training set of sample raw scans. The set may include in-class objects and not-in-class raw scans. An AI classifier may be configured to sample raw scans in the training set, measure errors in the results, update classifier parameters based on the errors, and detect completion of training.