Kernel Coherence Pursuit: A Manifold Learning-based Outlier Detection Technique

An efficient outlier detection method for manifold structures, termed Non-linear Coherence pursuit (NCoP), is presented. Inliers are assumed to be samples from a low-dimensional non-linear manifold, and thus should be highly coherent. Since outliers typically do not follow this structure, they are l...

Full description

Saved in:
Bibliographic Details
Main Authors: Sedghi, Mahlagha, Atia, George, Georgiopoulos, Michael
Format: Conference Proceeding
Language:English
Subjects:
Anomaly detection
Coherence
Data models
Iterative methods
Kernel
Manifolds
Principal component analysis
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:An efficient outlier detection method for manifold structures, termed Non-linear Coherence pursuit (NCoP), is presented. Inliers are assumed to be samples from a low-dimensional non-linear manifold, and thus should be highly coherent. Since outliers typically do not follow this structure, they are likely to have low mutual coherence with a large number of data points. A tailored notion of mutual coherence is introduced, and outliers are detected from inliers by comparing their total coherency values. We propose to utilize a well-suited measure of similarity using a radial basis kernel to capture nonlinear behaviors effectively. NCoP is remarkably simple, does not involve iterative algorithms, bears high ratios of outliers to inliers, and outperforms the state-of-the-art outlier detection methods for non-linear data patterns.
ISSN:2576-2303
DOI:10.1109/ACSSC.2018.8645334