Hierarchical Spiking Neural Networks for Contextual Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection methods lack a contextual framework and effective comparison metric for identifying anomalies in complex systems, particularly in streaming data and genomic edits, where spatial and temporal contexts are crucial.
Innovation Solution
A spiking neural network approach that utilizes phase-coded spikes, calculates median absolute differences within neighborhoods of varying sizes, and applies adaptive median filtering to detect anomalies by comparing input values to predefined thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional anomaly detection methods are used, then the detection process is simple, but they lack contextual framework and effective comparison metric for identifying anomalies in complex systems
Solution Approach 1:
The patent segments the anomaly detection process into multiple hierarchical levels (local neighborhoods, regional neighborhoods, global neighborhoods) with each level analyzing different spatial scales. This segmentation enables the system to capture contextual framework at multiple resolutions, improving detection accuracy while maintaining manageable complexity through modular processing stages.
Solution Approach 2:
The patent introduces spatial dimensionality by organizing data into hierarchical neighborhood structures (local, regional, global) that capture spatial context. This dimensional transformation from simple point-wise comparison to multi-scale spatial analysis provides the missing contextual framework and comparison metrics for complex systems while structured processing maintains computational feasibility.
2Measurement precision
If hierarchical neighborhood analysis is implemented, then detection accuracy is enhanced, but computational complexity increases
Solution Approach 1:
The computational workload is segmented across three hierarchical levels (local, regional, global neighborhoods), where each level processes data at its specific spatial scale. This segmentation distributes computational complexity across manageable stages rather than requiring simultaneous processing of all data points at the highest resolution, making the enhanced accuracy achievable through structured multi-scale analysis.
Solution Approach 2:
The patent applies partial action by focusing computational resources on local neighborhoods first to identify potential anomalies, then selectively expanding analysis to regional and global levels only where needed. This hierarchical filtering approach achieves high detection accuracy by concentrating detailed analysis on suspicious regions while using coarser scales for broader context, avoiding unnecessary computation on normal data points.
3Reliability
If multiple neighborhood sizes are analyzed, then false positives are reduced, but processing time increases
Solution Approach 1:
The patent performs preliminary analysis at the local neighborhood level first, establishing baseline anomaly indicators before expanding to larger regional and global neighborhoods. This preliminary action at finer scales filters out obvious anomalies early, reducing false positives by comparing against multiple spatial contexts while minimizing processing time through progressive rather than simultaneous multi-scale analysis.
Solution Approach 2:
The system applies partial action by analyzing multiple neighborhood sizes hierarchically rather than processing all scales equally. Local neighborhoods provide rapid initial filtering, regional neighborhoods refine detection with broader context, and global neighborhoods provide final verification. This selective progressive analysis reduces false positives through multi-context comparison while controlling processing time by avoiding redundant full-scale analysis of all data points.
Data Source
AI summary
Anomaly detection for streaming data is provided. A spiking neural network receives inputs of streaming data, wherein each input is contained within a number of neighborhoods and converts the inputs into phase-coded spikes. A median value of each input is calculated for each size neighborhood containing the input, and an absolute difference of each input from its median value is calculated for each size neighborhood. From the absolute differences, a median absolute difference (MAD) value of each input is calculated for each size neighborhood. It is determined whether the MAD value for any size neighborhood exceeds a respective threshold. If the MAD value exceeds its threshold, an anomaly indication is output for the input. If none of the MAD values for the neighborhoods exceeds its threshold, a normal indication is output for the input.