This week, Professor Andrea Cremaschi participated in the 18th International Joint Conference on Computational and Financial Econometrics (CFE) and Computational and Methodological Statistics (CMStatistics) at King’s College London. This renowned conference features presentations on computational and financial econometrics, as well as topics within the scope of the ERCIM Working Group CMStatistics.

Professor Cremaschi presented his innovative work titled “Matrix-variate priors for flexible mixture modelling of grouped data”. Over the past two decades, significant advancements have been made in Bayesian nonparametric literature, particularly in developing novel dependent prior distributions. These priors capture dependencies in heterogeneous data settings with grouped data, moving beyond standard univariate species sampling processes.

While much focus has been on nonparametric hierarchical processes like the Hierarchical Dirichlet Process, finite-dimensional dependent mixture models have received less attention. Professor Cremaschi’s work introduces a tractable class of dependent priors for mixture modelling using finite-dimensional matrix-variate distributions for the mixture weights. Specifically, the matrix-variate Dirichlet distribution is employed as a joint prior, ensuring positivity and the sum-to-one property while inducing dependence and borrowing information across groups.

This approach allows for flexible modelling of different data types and varying levels of data feature description, accommodating group-specific kernels. The proposed model is widely applicable and yields interpretable results. A tailored MCMC algorithm for posterior sampling was developed and demonstrated on both simulated and real-data examples.