Anomaly Cause Estimation Using Vector Likelihood Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection algorithms can determine the presence of anomalies but fail to provide detailed information on the anomaly itself.
Innovation Solution
An apparatus and method that utilize an input unit to process anomaly data detected by an anomaly detecting algorithm, searching for vectors that decrease the anomaly degree by considering the likelihood of each dimension being a cause, and estimating the anomaly cause based on these vectors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If anomaly detection algorithms output only anomaly degree, then the algorithm complexity is low and processing is simple, but detailed information on the anomaly cannot be obtained
Solution Approach 1:
The patent segments the anomaly analysis into two distinct stages: first, an anomaly detection algorithm outputs an anomaly degree to identify abnormal data; second, a separate cause estimation unit analyzes the anomaly data in detail to identify specific cause dimensions. This segmentation allows the system to maintain simple anomaly detection while adding detailed analysis only when needed, thus resolving the contradiction between algorithm simplicity and information completeness.
Solution Approach 2:
The patent performs preliminary anomaly detection using a simple algorithm to identify candidate anomaly data, then applies more complex cause estimation only to those identified anomalies. This preliminary action approach allows the system to maintain low overall complexity while obtaining detailed anomaly information for specific cases, rather than applying complex analysis to all data.
2Loss of information
If detailed cause analysis is performed for all data, then detailed anomaly information can be obtained, but the processing time and computational cost increase significantly
Solution Approach 1:
The patent applies preliminary anomaly detection to quickly identify which data points are abnormal, then performs detailed cause estimation only on those identified anomalies. This two-stage approach significantly reduces processing time compared to analyzing all data, while still obtaining detailed cause information for the anomalies that matter.
Solution Approach 2:
The patent performs cause estimation only partially - specifically for data identified as anomalies - rather than excessively analyzing all data. This partial action approach obtains necessary cause information for anomaly cases while avoiding the time cost of analyzing normal data, thus resolving the contradiction between information quality and processing efficiency.
3Device complexity
If anomaly detection is performed without considering dimension likelihood, then the detection process is simple, but the accuracy of cause identification is reduced
Solution Approach 1:
The patent applies local quality by introducing dimension likelihood specifically in the cause estimation stage for identified anomalies, rather than uniformly across all data processing. The cause estimation unit calculates likelihood for each dimension to determine which dimensions most likely caused the anomaly, providing accurate cause identification where needed while maintaining simplicity in the overall detection process.
Solution Approach 2:
The patent performs preliminary anomaly detection without dimension likelihood to identify candidate anomalies, then applies dimension likelihood analysis in a subsequent cause estimation stage. This preliminary action approach maintains detection simplicity while improving cause identification accuracy for the anomalies that are detected, resolving the contradiction between process simplicity and identification accuracy.
Data Source
AI summary
An apparatus for estimating anomaly information includes an input unit configured to input anomaly data detected as anomaly by an anomaly detecting algorithm that outputs an anomaly degree of input data for vectors, using learning of the vectors in a normal state, and an estimate unit configured to search for one or more vectors that decrease the degree of anomaly when added to the anomaly data, taking into account a likelihood, for each dimension, of a given dimension being a cause of the anomaly, and estimate the cause of the anomaly based on the searched vectors whereby it is possible to estimate detailed information on a detected anomaly.