Mario Beraha

I am a researcher (RTDa) at the Department of Mathematics of the Polytechnic University of Milan.

My research focus is broadly in the intersection of Bayesian nonparametrics and Machine Learning. More in detail I’m currently interested in discrete random measures, computational methods (MCMC and ABC), optimal transport and Wasserstein metrics, statistical modelling.

Reasonably up to date CV

Education

PhD in Data Science and Computation Università di Bologna and Politecnico di Milano (2018 - 2022)

MSc in Mathematical Engineering Politecnico di Milano (2015-2018)

BSc in Mathematical Engineering Politecnico di Milano (2012-2015)

Research

Submitted and ongoing projects:

Improved prediction of future user activity in online A/B testing (with L. Masoero, S. Favaro, and T. Richardson) [submitted, arXiv]
A Nonparametric Bayes Approach to Online Activity Prediction (with L. Masoero, S. Favaro, and T. Richardson) [submitted, arXiv]
Bayesian nonparametric boundary detection for income areal data (with M. Gianella and A. Guglielmi) [submitted, arXiv]
MCMC for Bayesian nonparametric mixture modeling under differential privacy (with S. Favaro and V. Rao) [submitted, arXiv]
Frequency and cardinality recovery from sketched data: a novel approach bridging Bayesian and frequentist views (with S. Favaro and M. Sesia) [submitted, arXiv]
Wasserstein Principal Component Analysis for circular measures (with M. Pegoraro) [arXiv, comments welcome!]
Transform-scaled process priors for trait allocations in Bayesian nonparametrics (with S. Favaro) [submitted, arXiv]
Random measure priors in Bayesian frequency recovery from sketches (with S. Favaro) [submitted, arXiv]
Bayesian clustering of high-dimensional data via latent repulsive mixtures (with L. Ghilotti and A. Guglielmi) [submitted, arXiv]
Normalized random measures with interacting atoms for Bayesian nonparametric mixtures (with R. Argiento, F. Camerlenghi and A. Guglielmi) [submitted, arXiv]
BayesMix: Bayesian mixture models in C++ [submitted, arXiv]

Publications in journals and international conferences

Bayesian nonparametric model based clustering with intractable distributions: an ABC approach (with R. Corradin) [Bayesian Analysis, arXiv]
Childhood Obesity in Singapore: a Bayesian Nonparametric Approach (with A. Guglielmi, F.A. Quintana, M. de Iorio, J.G. Eriksson, F. Yap)
[Statistical Modelling (2023), arXiv]
Normalized latent factor measure models (with J. E. Griffin) [Journal Royal Statistical Society Series B (2023), arXiv]
Projected Statistical Methods for Distributional Data on the Real Line with the Wasserstein Metric (with M. Pegoraro)
[Journal of Machine Learning Research (2022), arXiv]
Spatially dependent mixture models via the Logistic Multivariate CAR prior (with M. Pegoraro, R. Peli and A. Guglielmi)
[Spatial Statisitcs (2021), arXiv]
MCMC computations for Bayesian mixture models using repulsive point processes (with R. Argiento, J. Møller and A. Guglielmi)
[Journal of Computational and Graphical Statisics (2021), arXiv]
The semi-hierarchical Dirichlet Process and its application to clustering homogeneous distributions (with A. Guglielmi and F. A. Quintana)
[Bayesian Analysis (2021), arXiv]
Fast PCA in 1-D Wasserstein spaces via B-splines representation and metric projection with M. Pegoraro)
[AAAI'21, video]
Invited discussion on: "Latent nested nonparametric priors" by Camerlenghi et al. (with Alessandra Guglielmi)
[Bayesian Analysis (2019)]
Feature Selection via Mutual Information: New Theoretical Insights (with A. M. Metelli, M. Papini, A. Tirizoni and M. Restelli)
[IJCNN 2019, arXiv]

Other publications

JAGS, NIMBLE, Stan: a detailed comparison among Bayesian MCMC software (with D. Falco and A.Guglielmi) [arXiv]
Spatially dependent mixture models with a random number of components (with M. Gianella and A. Guglielmi) [SIS 2021]
Anisotropic determinantal point processes andtheir application in Bayesian mixtures (with L. Ghilotti and A. Guglielmi) [SIS 2021]
An ABC algorithm for random partitions arising from the Dirichlet process (with R. Corradin) [SIS 2020]
A Bayesian approach for modelling dependence among mixture densities (with M. Pegoraro, R. Peli and A. Guglielmi) [SIS 2020]
Choosing the right tool for the job: a systematic analysis of general purpose MCMC software (with G. Gualtieri, E. Villa, R. Vitali and A. Guglielmi) [SIS 2020]
A Bayesian model for network flow data: an application to BikeMi trips (with G. Bissoli, C. Principi, G. M. Rinaldi and A. Guglielmi) [SIS 2019]

Talks and Seminars

Learning to (approximately) count using Bayesian nonparametrics (invited talk) [CMStat 2023, Dec. 2023]
A BNP approach to online user activity prediction [BAYSM, Nov. 2023]
Random measure priors in Bayesian frequency recovery from sketches (invited talk) [EcoStat 2023, Aug. 2023]
Random measure priors in Bayesian frequency recovery from sketches (invited seminar) [School of Mathematical Sciences.University of Nottingham, March 2023]
Beyond CRMs: normalized random measures with atoms' interactions (invited talk) [ISBA 2022, June 2022]
Beyond CRMs: normalized random measures with atoms' interactions (invited talk) [3rd Italian meeting on probability and mathematical statistics, June 2022]
Distributional data analysis with the Wasserstein distance (invited seminar) [HDFD group, UCL, May 2022]
Normalized Latent Factor Measure Models (invited talk) [BNP Networking in Cyprus, Apr. 2022]
Repulsive mixture models: modelling and computations, with applications to high-dimensional data (invited talk) [CMStatistics, Dec. 2021]
MCMC computations for Bayesian mixture models using repulsive point processes (invited talk) [CLADAG, Sept. 2021]
The semi-hierarchical Dirichlet Process and its application to clustering homogeneous distributions (contributed talk) [ISBA 2021]
Anisotropic determinantal point processes and their application in Bayesian mixtures (contributed talk) [SIS 2021]
Projected statistical methods in the Wasserstein space (department seminar) [MOX - Politecnico di Milano, Jan 2021]
MCMC computations for Bayesian mixture models using repulsive point processes (invited talk) [CMStatistics, Dec. 2020]
MCMC computations for Bayesian mixture models using repulsive point processes (invited talk) [BAYSM:O, Nov. 2020 ]
Projected statistical methods in the Wasserstein space (Invited Seminar) [Inria Grenoble, team Statify, Nov. 2020]
Spatially dependent mixture models via the Logistic Multivariate CAR prior (prerecorded talk) [Bernoulli-IMS One World Symposium 2020]
Invited talk at CompStat '2020 (Cancelled)
Invited discussion on: "Latent nested nonparametric priors" by Camerlenghi et al. (talk) [Webinar]
Zero-inflated Poisson regression models for the analysis of network flow data (invited talk) [ARS 2019, Salerno]
Feature Selection via Mutual Information: New Theoretical Insights (talk) [IJCNN 2019, Budapest]
Bayesian Nonparametric Vector Autoregressive models (poster) [BNP 12, Oxford]
A Bayesian model for network flow data: an application to BikeMi trips (talk) [SIS 2019, Milano]