Kernel Two-Sample and Independence Tests for Nonstationary Random Processes

Two-sample and independence tests with the kernel-based mmd and hsic have shown remarkable results on i.i.d. data and stationary random processes. However, these statistics are not directly applicable to nonstationary random processes, a prevalent form of data in many scientific disciplines. In this work, we extend the application of mmd and hsic to nonstationary settings by assuming access to independent realisations of the underlying random process. These realisations—in the form of nonstationary time-series measured on the same temporal grid—can then be viewed as i.i.d. samples from a multivariate probability distribution, to which mmd and hsic can be applied. We further show how to choose suitable kernels over these high-dimensional spaces by maximising the estimated test power with respect to the kernel hyperparameters. In experiments on synthetic data, we demonstrate superior performance of our proposed approaches in terms of test power when compared to current state-of-the-art functional or multivariate two-sample and independence tests. Finally, we employ our methods on a real socioeconomic dataset as an example application.

Generalization of Spectral Methods for High-Cycle Fatigue Analysis to Accommodate Non-Stationary Random Processes

Estimation of material fatigue life is an essential task in many engineering fields. When non-proportional loads are applied, the methodology to estimate fatigue life grows in complexity. Many methods have been proposed to solve this problem both in time and frequency domains. The former tends to give more accurate results, while the latter seems to be more computationally favorable. Until now, the focus of frequency-based methods has been limited to signals assumed to follow a stationary statistic process. This work proposes a generalization to the existing methods to accommodate non-stationary processes as well. A sensitivity analysis is conducted on the influence of the formulation’s hyper-parameters, followed by a numerical investigation on different signals and various materials to assert the robustness of the method.

Methods of mathematical modeling of non-Gaussian random variables quantities and processes

Mathematical methods allowing to model non-Gaussian random variables and processes are considered. The models and description of non-Gaussian correlated processes in the form of generated Gaussian noise are analyzed, as well as the methods of formation of stationary random processes defined by the one-dimensional density distribution of Vero -abilities and the autocorrelation function. Examples of formation of non-Gaussian random variables and processes are given.

Construction of estimates of spectral densities with a given accuracy over intersecting intervals of observations

The article proposes a new method for determining the number of splitting intervals and the number of observations in them when building estimates of the spectral densities of stationary random processes with a given accuracy over intersecting observation intervals based on asymptotic results, obtained for the first moment of convergence rate under the assumption that the spectral density satisfies the Lipschitz condition. Two cases are considered: with a single and arbitrary data taper. As a result, an algorithm is proposed for constructing estimates for intersecting intervals of observations with a given accuracy. This algorithm was tested on model examples for random AR(4) processes, using data taper of Riesz, Bochner, Parzen. The proposed method will be useful to the researcher in analyzing data in the form of stationary random processes using non­parametric methods of spectral analysis in an automated mode.

Ornstein Auto-Encoders

We propose the Ornstein auto-encoder (OAE), a representation learning model for correlated data. In many interesting applications, data have nested structures. Examples include the VGGFace and MNIST datasets. We view such data consist of i.i.d. copies of a stationary random process, and seek a latent space representation of the observed sequences. This viewpoint necessitates a distance measure between two random processes. We propose to use Orstein's d-bar distance, a process extension of Wasserstein's distance. We first show that the theorem by Bousquet et al. (2017) for Wasserstein auto-encoders extends to stationary random processes. This result, however, requires both encoder and decoder to map an entire sequence to another. We then show that, when exchangeability within a process, valid for VGGFace and MNIST, is assumed, these maps reduce to univariate ones, resulting in a much simpler, tractable optimization problem. Our experiments show that OAEs successfully separate individual sequences in the latent space, and can generate new variations of unknown, as well as known, identity. The latter has not been possible with other existing methods.