IE Research Datalab member Andrea Cremaschi presented the seminar “Dependent Dirichlet-Multinomial Processes with Random Number of Components” at Collegio Carlo Alberto in Turin on December 5, 2025.

The talk introduced a new class of finite Bayesian nonparametric models for dependent data. The approach places a prior on the number of mixture components and uses correlated Wishart-based weights to represent different levels of dependence while retaining useful theoretical properties and efficient posterior computation.

Cremaschi demonstrated the models through simulation studies and two real-data applications, including an analysis of sex-specific differences in gene expression in the human brain. The results showed how the proposed processes can flexibly capture complex dependence structures without relying on computationally costly variable-augmentation schemes.

The seminar forms part of Cremaschi’s research in Bayesian statistics, mixture modelling, clustering, and applications to biomedical data.