System and Method for Combining Learning Systems

Simple SummaryContent extracted from patent full text and abstract with AI.

This patent describes a system and method for combining at least two different machine learning (ML) systems to perform data processing, specifically for classifying system states (such as normal or abnormal). It processes measurement data using both time domain and frequency domain ML models, each generating its own score. The results from these models are then grouped and merged via specialized functions to produce an overall state classification. The system can dynamically adapt its configuration based on previous classification results to improve accuracy.

Use CasesContent extracted from patent full text and abstract with AI.

• Real-time monitoring of IT systems for anomaly detection (e.g., detecting security breaches or system failures)
• Industrial process control to identify equipment malfunctions or abnormal operation
• Telecommunications network monitoring to detect performance issues or unusual patterns
• Healthcare monitoring systems (e.g., patient vital signs, detecting abnormal trends)
• Financial transaction analysis to detect fraud or unusual activities
• Smart building management (e.g., monitoring HVAC or energy usage patterns)

BenefitsContent extracted from patent full text and abstract with AI.

• Improved classification accuracy by leveraging both time and frequency domain analyses, capturing a wider range of anomalies
• Reduced false positive and false negative rates compared to using a single ML system
• Dynamic adaptation of learning models based on current and past classification results, enabling continuous improvement
• Applicable to varied types of data and different domains requiring anomaly or state detection
• Flexible merging and consolidation functions allow customization to specific monitoring or detection requirements

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

A method for combing least two learning systems for data processing, wherein measurement result data of at least one predetermined time interval is measured and provided to at least two learning systems, wherein each measurement result data of a predetermined time interval is separately processed by each of the learning systems and the learning systems are configured to calculate a respective numeric score for each measurement and each learning system, wherein the respective numeric score is classified by a respective consolidation function and the consolidation function is configured to classify a plurality of respective numeric scores into at least two respective classification groups, each classification group corresponding to at least two states, and wherein the respective classification groups of the at least two learning systems are merged by a classification group merging function to at least one state classification based on a merging model.