Time series on riemannian manifolds

Download
2017
Ergezer, Hamza
In this thesis, feature covariance matrices are utilized to solve several problems related to time series. In the first part of the thesis, a novel representation is proposed to represent the time series using feature covariance matrices. By this representation, time series are carried onto Riemannian manifold space. The proposed representation is firstly applied to trajectories which are essentially 2D time series. Anomaly detection and activity perception problems in crowded visual scenes are studied by using the trajectories. The second utilization of the proposed representation is for classification of 1D time series. The feature covariance matrices of overlapping subsequences are extracted and fed into two well-known classifiers as the input. The last contribution of the thesis is a rank-based distance measure for high dimensional covariance matrices. The distance measure is utilized to solve skeletal action recognition problem. Unlike classical distance measures, the rank-based distance measure enables us to learn the manifold structure. For this reason, essentially, it can be asserted that the proposed approach is about manifold learning. Performances of the approaches proposed in this thesis have been compared to most of the state-of-the-art techniques on publicly available well-known datasets. For all of the studied problems, we achieve comparable or outperforming results compared to the state-of-the-art techniques.

Suggestions

Time series classification with feature covariance matrices
Ergezer, Hamza; Leblebicioğlu, Mehmet Kemal (2018-06-01)
In this work, a novel approach utilizing feature covariance matrices is proposed for time series classification. In order to adapt the feature covariance matrices into time series classification problem, a feature vector is defined for each point in a time series. The feature vector comprises local and global information such as value, derivative, rank, deviation from the mean, the time index of the point and cumulative sum up to the point. Extracted feature vectors for the time instances are concatenated t...
Time Series Classification Using Point-wise Features
Ergezer, Hamza; Leblebicioğlu, Mehmet Kemal (2017-05-18)
In this work, a novel approach utilizing feature covariance matrices is proposed for time series classification. In order to adapt the feature covariance matrices for time series classification, a feature vector is defined for each point in a time series. The feature vector comprises local and global information such as value, derivative, rank, deviation from the mean, time index of the point and cumulative sum up to the point. Instead of representing the whole time series with a single covariance matrix, t...
STABILITY OF CONTROL FORCES IN REDUNDANT MULTIBODY SYSTEMS
IDER, SK (1996-01-03)
In this paper inverse dynamics of redundant multibody systems using a minimum number of control forces is formulated. It is shown that the control forces and the task accelerations may become noncausal at certain configurations, yielding the dynamical equation set of the system to be singular. For a given set of tasks, each different set of actuators leads to a different system motion and also to different singular configurations. To avoid the singularities in the numerical solution, the dynamical equations...
Distance-based discretization of parametric signal manifolds
Vural, Elif (2010-06-28)
The characterization of signals and images in manifolds often lead to efficient dimensionality reduction algorithms based on manifold distance computation for analysis or classification tasks. We propose in this paper a method for the discretization of signal manifolds given in a parametric form. We present an iterative algorithm for the selection of samples on the manifold that permits to minimize the average error in the manifold distance computation. Experimental results with image appearance manifolds d...
Frequency estimation of a single real-valued sinusoid: An invariant function approach
Candan, Çağatay; Çelebi, Utku (2021-08-01)
An invariant function approach for the computationally efficient (non-iterative and gridless) maximum likelihood (ML) estimation of unknown parameters is applied on the real-valued sinusoid frequency estimation problem. The main attraction point of the approach is its potential to yield a ML-like performance at a significantly reduced computational load with respect to conventional ML estimator that requires repeated evaluation of an objective function or numerical search routines. The numerical results ind...
Citation Formats
H. Ergezer, “Time series on riemannian manifolds,” Ph.D. - Doctoral Program, Middle East Technical University, 2017.