The Institute for Statistics and Mathematics is pleased to invite you to the next research seminar, taking place on campus:

Alejandra Avalos Pacheco (Institute of Statistics and Mathematical Methods in Economics, TU Wien;
Harvard-MIT Center for Regulatory Science, Harvard University):
Integrative Large-Scale Bayesian Learning: From Factor Analysis to Graphical Models
Friday, May 5, 2023, 8:30 am, Room D4.4.008

Abstract: Data integration is crucial when separate data sources are associated on the same phenomenon. Integrative models provide gains in statistical power and help to take accurate decisions sooner. However, the lack of proper integration tools could lead to unreliable and misleading inference.
In this talk I will present novel methods to integrate continuous and binary heterogeneous data: sparse factor regression (FR), multi-study factor regression (MSFR) and multiple Ising graphical (MIG) models. The FR model provides a tool for continuous data exploration via dimensionality reduction and sparse low-rank covariance estimation, while correcting for a range of systematic biases. MSFR models are extensions of FR that enable us to jointly obtain a covariance structure that identify and estimate the group-specific covariances in addition to the common components. The MIG model studies the heterogeneity induced in a set of binary variables by external factors and provides the embedded network structures of distinct groups.
I will discuss the use of several priors that lead to more interpretable models, such as sparse priors (local and non-local), which learn the dimension of the latent factors; and Markov Random Field priors, which enable the borrowing of strength across different groups.
Finally, I will show the usefulness of our methods by providing a visual representation of the data and by answering data-specific questions, such as: providing survival predictions in different cancer patients, associating cardio-metabolic disease risks with dietary patterns of distinct latino populations, or analysing the confidence in political institutions in different web engagement segments groups. 
I will highlight the benefits of our models compared to other techniques, not only in the accuracy of the data signals but also in a better prediction, and I will provide computationally efficient tools for inference.

We aim to stream all on-campus talks via Zoom. A direct link to the stream will be posted on our website.

For further information and the seminar schedule, please see:
www.wu.ac.at/en/statmath/research/resseminar