Anomaly Cause Extraction for Variable-Length Communication Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection technologies using deep learning fail to specify the causes of abnormalities in communication data with variable lengths.
Innovation Solution
An estimation device that uses dynamic programming to calculate similarity between abnormal and normal data, identifying a portion with no correspondence as the cause of abnormality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning is used for anomaly detection, then detection accuracy is improved, but the ability to specify causes of abnormality deteriorates due to the black box nature of deep learning
Solution Approach 1:
The patent introduces an intermediary component that acts as a bridge between the deep learning anomaly detection system and the cause specification requirement. This intermediary processes the output of the deep learning model and maps it back to specific portions of the input data, thereby recovering the lost interpretability without compromising the detection accuracy provided by the deep learning model.
2Loss of information
If communication data with fixed lengths is used as abnormality detection target, then cause specification is enabled, but the ability to handle variable length communication data deteriorates
Solution Approach 1:
The patent applies dynamics by making the data structure adaptable rather than fixed. The system dynamically adjusts to handle communication data of variable lengths while maintaining the capability to specify causes of abnormalities. This is achieved through a flexible mapping mechanism that can accommodate different data lengths without requiring fixed-length constraints.
Data Source
Figure 1(a)~1(c)
Figure 2
Figure 3
AI summary
An estimation unit (15b) estimates normal data having maximum similarity between abnormal data determined to be abnormal and each of a plurality of pieces of normal data determined to be normal. An extraction unit (15c) compares abnormal data with normal data estimated to have the maximum similarity and extracts a portion of the abnormal data having no portion corresponding to the normal data as a cause portion of the abnormality.