Anomaly Detection Using Adaptive Spatial Voting Grid
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Devices with limited memory resources face challenges in efficiently detecting anomalous behavior due to high memory usage requirements of existing anomaly detection methods, which are not adaptive and require extensive recalculations of table indices.
Innovation Solution
The implementation of a self-adaptive Spatial Voting (SV) grid using Run Length Encoding (RLE) to represent the SV grid, allowing for efficient detection of anomalies in a fixed amount of memory, with adaptive partitioning and encoding of data to reduce memory consumption and processing cycles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional anomaly detection methods are used, then detection accuracy is improved, but memory usage increases significantly
Solution Approach 1:
The patent segments the anomaly detection process into two distinct phases: an offline training phase that builds anomaly profiles and signatures, and an online detection phase that uses these pre-computed profiles with minimal memory requirements. This segmentation allows complex analysis to be performed when memory is available, while runtime operation uses compact representations.
Solution Approach 2:
The patent performs preliminary action by pre-computing anomaly profiles, signatures, and baseline behaviors during an offline training phase. These pre-computed structures are stored and then used during online detection without requiring extensive recalculations, thereby reducing runtime memory usage while maintaining detection accuracy.
2Adaptability or versatility
If adaptive anomaly detection is implemented, then detection flexibility is improved, but processing complexity increases
Solution Approach 1:
The patent implements dynamics by creating adaptive anomaly detection systems that can adjust to new patterns and behaviors. The system dynamically updates anomaly profiles and signatures based on learned patterns, allowing it to adapt to changing conditions while using structured approaches to manage processing complexity.
Solution Approach 2:
The patent applies self-service through self-training mechanisms where the system automatically learns from data, builds its own anomaly profiles, and improves its detection capabilities without requiring extensive manual configuration or external intervention, thereby managing complexity through automation.
3Measurement precision
If extensive training data is used, then model accuracy is improved, but training time and resource requirements increase
Solution Approach 1:
The patent performs preliminary action by conducting comprehensive training and data analysis in an offline phase before deployment. During this offline training phase, the system processes extensive training data to build anomaly profiles and signatures. Once trained, the system can operate with minimal memory and processing requirements during online detection, effectively shifting the time and resource burden to the offline phase.
Data Source
AI summary
A method can include identifying, by processing circuitry of a device, a row of rows and/or column of columns to which a first feature and a second feature of an input maps, comparing the identified row and/or column to a row run length encoding (RLE) in a memory of the device that indicates, for each row, whether one or more cells in the row include an input mapped thereto or a column RLE in a memory of the device that indicates, for each column, whether one or more cells in the column include an input mapped thereto, respectively, and determining the input data is anomalous in response to determining either the row RLE indicates that no inputs are mapped to the row to which the input maps, or the column RLE indicates that no inputs are mapped to the column to which the input maps.