Stationarity of Time-Series on Graph via Bivariate Translation Invariance
Stationarity is a cornerstone in classical signal processing (CSP) for modeling and characterizing various stochastic signals for the ensuing analysis. However, in many complex real world scenarios, where the stochastic process lies over an irregular graph structure, CSP discards the underlying structure in analyzing such structured data. Then it is essential to establish a new framework to analyze the high-dimensional graph structured stochastic signals by taking the underlying structure into account. To this end, looking through the lens of operator theory, we first propose a new bivariate isometric joint translation operator (JTO) consistent with the structural characteristic of translation operators in other signal domains. Moreover, we characterize time-vertex filtering based on the proposed JTO. Thereupon, we put forth a new definition of joint wide-sense stationary (JWSS) signals in time-vertex domain using the proposed isometric JTO together with its spectral characterization. Then a new joint power spectral density (JPSD) estimator, called generalized Welch method (GWM), is presented. Simulation results are provided to show the efficacy of this JPSD estimator. Furthermore, to show the usefulness of JWSS modeling, we focus on the classification of time-series on graph. To that end, by modeling the brain Electroencephalography (EEG) signals as JWSS processes, we use JPSD as the feature for the Emotion and Alzheimer's disease (AD) recognition. Experimental results demonstrate that JPSD yields superior Emotion and AD recognition accuracy in comparison with the classical power spectral density (PSD) and graph PSD (GPSD) as the feature set for both applications. Eventually, we provide some concluding remarks.
Code (0)
등록된 구현이 없습니다.
Tasks
EEGElectroencephalogram (EEG)Time SeriesTime Series AnalysisTranslationSimilar Papers 제목 키워드 기반
Causal Inference from Slowly Varying Nonstationary Processes
Causal inference from observational data following the restricted structural causal model (SCM) framework hinges largely on the asymmetry between cause and effect from the data generating mechanisms, such as non-Gaussian…
Causal DiscoveryCausal IdentificationCausal InferenceTime Series+1Causal Inference from Slowly Varying Nonstationary Processes
Causal inference from observational data following the restricted structural causal models (SCM) framework hinges largely on the asymmetry between cause and effect from the data generating mechanisms, such as non-Gaussia…
Causal IdentificationCausal InferenceTime SeriesDetrended partial cross-correlation analysis of two nonstationary time series influenced by common external forces
When common factors strongly influence two power-law cross-correlated time series recorded in complex natural or social systems, using classic detrended cross-correlation analysis (DCCA) without considering these common …
Time SeriesTime Series AnalysisCausal Discovery with General Non-Linear Relationships Using Non-Linear ICA
We consider the problem of inferring causal relationships between two or more passively observed variables. While the problem of such causal discovery has been extensively studied especially in the bivariate setting, the…
Causal DiscoveryDeep reinforcement learning for time series: playing idealized trading games
Deep Q-learning is investigated as an end-to-end solution to estimate the optimal strategies for acting on time series input. Experiments are conducted on two idealized trading games. 1) Univariate: the only input is a w…
Deep Reinforcement LearningQ-Learningreinforcement-learningReinforcement Learning+3