Localized Anomaly Detection Models Without Global Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Converting global anomaly detection models to localized spaces is time-consuming and difficult, as existing methods require rebuilding models from scratch for each localized space, leading to inefficiencies and management challenges.
Innovation Solution
A transformation method that employs a cost function to map a global anomaly detection model to a localized space by estimating a scaling factor using global vectors, allowing the generation of a local anomaly detection model without re-training, thus reducing complexity and processing burden.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a global anomaly detection model is converted to localized spaces by rebuilding from scratch, then the model can be tailored to specific geographic regions or entities, but the process becomes time-consuming and difficult to manage
Solution Approach 1:
The patent segments the global anomaly detection model into multiple localized models, each tailored to specific geographic regions or entities. Instead of rebuilding from scratch, the system extracts and transforms relevant portions of the global model for each local context, reducing redundant computation while maintaining customization.
Solution Approach 2:
The system performs preliminary actions by pre-processing and storing global model data in a structured format that can be quickly transformed for local use. The global model is prepared in advance with extracted features and patterns that can be efficiently reused across different localizations without requiring complete retraining.
2Reliability
If a global anomaly detection model is converted to localized spaces by rebuilding from scratch, then the model can be optimized for local data characteristics, but the complexity and management difficulty increase
Solution Approach 1:
The patent creates a universal framework that generates multiple localized models from a single global model. The same global model structure and training methodology can be applied across different geographic regions and entities, reducing management complexity while maintaining local optimization through parameter transformation rather than complete rebuilding.
3Measurement precision
If localized anomaly detection models are built from scratch, then the models can be fully optimized for local data, but the processing burden and computational resources increase
Solution Approach 1:
Instead of creating entirely new models from scratch for each local context, the system copies and transforms the global model structure for local use. This approach preserves the computational efficiency of the global model while adapting it to local data characteristics through parameter transformation, significantly reducing the processing burden and energy consumption compared to complete retraining.
Data Source
AI summary
A local anomaly detection model for monitoring a local entity set of a network system is generated by applying a transformation function to a global anomaly detection model for the network system, without re-training the global anomaly detection model for the local entity set. The global anomaly detection model, which may be generated via unsupervised learning methods, includes global vector(s) of metrics and a global healthy vector space. The transformation function is estimated and applied to the global anomaly detection model to generate the local anomaly detection model, which includes local vector(s) of metrics pertaining to the local entity set and a local healthy vector space. Responsive to a determination that the local vector(s) of metrics comprises one or more anomalous data points outside of the local healthy vector space, an alert regarding the one or more anomalous data points can be generated and output.