Abstract: Multivariate spatio-temporal data, characterized by complex dependencies across both spatial and temporal dimensions, present substantial challenges in modeling and prediction. Dimension reduction methods, particularly those based on blind source separation (BSS), offer a powerful framework for addressing these challenges, provided they can effectively capture the nonlinear and nonstationary nature of the data. In this talk, we suggest using identifiable variational autoencoders (iVAEs) tailored for nonlinear, nonstationary spatio-temporal BSS, for this purpose.

We further propose two strategies for estimating the number of the latent independent components, a critical aspect of accurate representation learning. The effectiveness of our method is demonstrated through extensive simulation studies and a real-world application.

Speaker:  Klaus Nordhausen is a Professor of Statistics at the University of Helsinki, specializing in multivariate statistical methods. His research includes supervised and unsupervised dimension reduction, blind source separation, and robust and nonparametric methods. He holds an MSc in Statistics from the University of Dortmund and a PhD in Biometry from the University of Tampere. He has held academic positions at the University of Turku, the Technical University of Vienna, and the University of Jyväskylä.

Affiliation: University of Helsinki

Place of Seminar:  Kumpula Exactum BK106 (in person) & zoom ( Meeting ID: 640 5738 7231 ; Passcode: 825217)