Inverse-Density Exemplar Selection for Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models for anomaly detection in time series data often perform poorly at extreme activity levels due to biased exemplar selection, favoring non-extreme vectors from the distribution's middle, leading to reduced accuracy in detecting anomalies at the edges of the demand profile.
Innovation Solution
The inverse-density exemplar selection system iteratively increases the proportion of extreme vectors from the distribution's edges and decreases non-extreme vectors from the middle, adjusting selection criteria to improve the machine learning model's performance by incorporating more representative data for extreme asset activity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional exemplar selection methods are used that favor non-extreme vectors from the distribution's middle, then the training process is simpler and faster, but the anomaly detection accuracy at extreme activity levels deteriorates
Solution Approach 1:
The patent inverts the traditional exemplar selection approach by prioritizing extreme vectors from the distribution's edges over non-extreme vectors from the middle. Instead of selecting exemplars based on density (which favors the middle), the method selects based on extremity, thereby improving anomaly detection accuracy at extreme activity levels while maintaining a manageable subset size through iterative selection with replacement.
2Measurement precision
If a larger proportion of extreme vectors is included in the exemplar set, then the detection accuracy for extreme activity improves, but the computational resources and training time increase
Solution Approach 1:
The patent applies partial action by selecting a subset of extreme vectors rather than using all available data. The iterative selection process with replacement ensures that the exemplar set contains a sufficient proportion of extreme vectors to improve detection accuracy, while maintaining a limited subset size that conserves computational resources and training time.
3Reliability
If the exemplar selection is biased toward non-extreme vectors, then the training data is more homogeneous and easier to process, but the false-alarm and missed-alarm probabilities increase at extreme activity levels
Solution Approach 1:
The patent applies local quality by tailoring the exemplar selection to the specific requirement of detecting anomalies at extreme activity levels. Instead of using a uniform selection approach across the entire distribution, the method selectively prioritizes extreme vectors from the distribution's edges, thereby improving reliability at extreme levels while maintaining overall processing efficiency through the iterative subset selection process.
Data Source
AI summary
Systems, methods, and other embodiments associated with inverse-density exemplar selection for improved multivariate anomaly detection are described. In one embodiment, a method includes determining magnitudes of vectors from a set of time series readings collected from a plurality of sensors. And, the example method includes selecting exemplar vectors from the set of time series readings to train a machine learning model to detect anomalies. The exemplar vectors are selected by repetitively (i) increasing a first density of extreme vectors that are within tails of a distribution of amplitudes for the time series readings based on the magnitudes of vectors, and (ii) decreasing a second density of non-extreme vectors that are within a head of the distribution based on the magnitudes of vectors. The repetition continues until the machine learning model generates residuals within a threshold in order to reduce false or missed detection of the extreme vectors as anomalous.