Anomaly Factor Estimation Using L2 Norm Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing algorithms for anomaly detection in multidimensional data struggle to correctly extract factor dimensions without assuming sparsity, especially when anomalies spread across many dimensions, leading to incorrect factor estimation.
Innovation Solution
An anomaly factor estimation apparatus is developed, comprising a training unit for anomaly detection using autoencoders, a testing unit for determining anomalies, and a factor estimation unit that calculates the degree of contribution for each dimension based on likelihood, allowing for accurate extraction of factor dimensions without sparsity assumptions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sparse optimization is used to estimate anomaly factors, then the method can search for input data items that make the degree of anomalousness smaller, but the factor estimation becomes incorrect when factors spread over many dimensions
Solution Approach 1:
The patent changes the optimization parameter from L0 norm (sparse optimization) to L2 norm (least squares optimization). This parameter change allows the system to handle cases where anomaly factors spread over many dimensions without assuming sparsity, thereby resolving the contradiction between factor estimation accuracy and adaptability to multi-dimensional factors.
2Productivity
If sparsity assumption is made for factor dimensions, then the method can extract factors more efficiently, but the extraction fails when anomalies span multiple dimensions
Solution Approach 1:
The patent changes the optimization criterion from sparse optimization (L0 norm) to least squares optimization (L2 norm). This parameter change removes the sparsity assumption, allowing the system to maintain both efficiency and correctness when extracting factors that span multiple dimensions, thereby resolving the contradiction between productivity and reliability.
Data Source
AI summary
An anomaly factor estimation apparatus includes a memory and a processor configured to execute training an anomaly detection model for performing anomaly detection based on an error of output data with respect to input data represented by a multidimensional numerical vector; inputting test data into the anomaly detection model, to determine whether the test data corresponds to an anomaly based on output data output from the anomaly detection model and the test data; and calculating a degree of contribution to the anomaly for each dimension of the test data based on a likelihood of said each dimension, in response to determining that the test data corresponds to the anomaly.