Virtual Seminar Series

Speaker: Tim Kutta, Aarhus University​

Title: Monitoring time series - what has changed?​

Abstract: ​​​​​​​ In this talk, we discuss the problem of sequential change point detection. In particular, we focus on sequentially testing for changes in the mean of a dependent time series relative to a training period. We consider approaches for scalar and functional data. We highlight that, while existing procedures are reliable, they often detect changes slowly, and little progress has been made on this front in the past 20 years. We then explore how the field might move forward and obtain more useful, fast and powerful procedures, based on recent advances in stochastic process theory.

Speaker: Anru Zhang, Duke University​

Title: Recent Advances in Generative Modelling and Synthetic Biomedical Data​

Abstract: The growing use of electronic health records (EHRs) and biomedical data presents new opportunities and challenges for developing generative models that can produce realistic synthetic data while safeguarding privacy and reducing bias. In this talk, I will present recent advances in generative modelling for biomedical data, including diffusion-based methods for privacy-preserving EHR time series, a new semiparametric copula flow approach for irregularly sampled functional data, and a bias-corrected synthesis framework that improves learning in rare-event settings. Together, these techniques form a principled toolkit for generating reliable, privacy-conscious synthetic data and expanding the usability of sensitive biomedical datasets. ​​​

Speaker: Francesco Sanna Passino, Imperial College London​

Title: On spectral embedding of dynamic multiplex graphs and subsequent inference tasks

Abstract: Many real-world networks evolve dynamically over time and present different types of connections between nodes, often called layers. In this talk, we propose and discuss a latent position model for these objects, called the dynamic multiplex random dot product graph (DMPRDPG), which uses an inner product between layer-specific and time-specific latent representations of the nodes to obtain edge probabilities. We further introduce a computationally efficient spectral embedding method for estimation of DMPRDPG parameters, called doubly unfolded adjacency spectral embedding (DUASE). The DUASE estimates are proved to be both consistent and asymptotically normally distributed. A key strength of our method is the encoding of time-specific node representations and layer-specific effects in separate latent spaces, which allows the model to capture complex behaviors while maintaining relatively low dimensionality. The embedding method we propose can also be efficiently used for subsequent inference tasks, such as clustering, anomaly detection and changepoint detection. In particular, we show how DUASE can be used to predict the response associated with an unlabeled time series of networks in a semisupervised setting, and discuss the use of DUASE for statistical hypothesis testing for differences between layers. Applications on real-world networks such as biological learning networks of larval Drosophila demonstrate practical uses of our method.​​​​​​​