Anomaly Detection Model Training with Semi-Supervised Labeling

The patent introduces a semi-supervised machine learning model as an intermediary between unsupervised anomaly detection and final supervised classification. This intermediate model pre-labels anomalies using uncertainty sampling, reducing the burden on manual analysis while maintaining accuracy. The semi-supervised model acts as a mediator that processes anomalies before they reach the supervised classification stage, thereby reducing time consumption without sacrificing detection precision.

The patent segments the anomaly detection process into three distinct stages: unsupervised detection, semi-supervised pre-labeling, and supervised classification. By dividing the workflow into these segments, each handling specific aspects of anomaly processing, the system achieves high accuracy while reducing manual intervention time. The segmentation allows automated processing to handle routine tasks while human experts focus only on critical cases.

The patent applies preliminary action by using unsupervised models to detect anomalies and semi-supervised models to pre-label them before supervised training. This preparatory work creates a enriched labeled dataset that can be used for efficient supervised model training. The preliminary detection and labeling stages ensure that when supervised training begins, the model receives high-quality labeled data, improving detection accuracy without requiring extensive manual labeling that would slow down the process.

The semi-supervised machine learning model serves as an intermediary that generates additional labeled training data from unlabeled anomalies. This intermediate labeling process expands the training dataset without requiring proportional increases in manual labeling effort, enabling faster supervised training while maintaining or improving detection accuracy through the use of uncertainty sampling and expert-in-the-loop validation.

The patent applies partial action by having human experts label only a subset of anomalies - specifically those with high uncertainty or critical importance - rather than all detected anomalies. The semi-supervised model handles the labeling of remaining anomalies automatically. This partial manual intervention maintains model accuracy while significantly reducing operational complexity and resource requirements compared to complete manual labeling.

The semi-supervised machine learning model performs self-service by automatically generating labels for anomalies using uncertainty sampling and existing labeled data. This self-labeling capability reduces dependency on manual expert intervention, simplifying operations while maintaining accuracy. The model serves itself by identifying its own knowledge gaps and selectively seeking human input only when necessary.

The patent segments the detection system into unsupervised and supervised components, each handling different aspects of anomaly detection. The unsupervised model provides automated initial detection, while the supervised model provides accurate classification. This segmentation allows the system to maintain high automation levels through the unsupervised component while achieving high accuracy through the supervised component, resolving the contradiction between automation and precision.

The semi-supervised machine learning model acts as an intermediary that bridges the gap between unsupervised detection and supervised classification. It takes the automatically detected anomalies from the unsupervised model and prepares them for supervised processing by pre-labeling based on uncertainty sampling. This intermediate step enables the system to maintain high automation while improving accuracy, as the semi-supervised model enhances the unsupervised detections without requiring full manual intervention.