Paul Egels (paul.egels@unibocconi.it)

I do research in mathematics applied to statistics and machine learning. Currently I am a postdoc reasearcher at Bocconi University working with Botond Szabo. Previously, I obtained my PhD from Sorbonne Université under the supervison of Ismaël Castillo.

My work focuses on high-dimensional statistics, in particular Bayesian nonparametrics, variational inference and Bayesian deep learning. I am also interested in approximation theory, adversarial robustness and other theoretical aspects of machine learning.

Papers & preprints

    Sergios Agapiou, Ismaël Castillo, Paul Egels. Leveraging p-exp tails for adaptation. preprint, 2026. [arXiv]

    Sergios Agapiou, Ismaël Castillo, Paul Egels. Heavy-tailed and Horseshoe priors for regression and sparse Besov rate. To appear in Bernoulli, 2026. [arXiv]

    Ismaël Castillo, Paul Egels. Posterior and variational inference for deep neural networks with heavy-tailed weights. Journal of Machine Learning Reaseach, 2025. [JMLR]

Working papers

    Ismaël Castillo, Paul Egels, Riccardo Lazzarini, Botond Szabo. Contraction rates of adversarially robust "pseudo"-posteriors. (Soon available).

    Julyan Arbel, Ismaël Castillo, Paul Egels. On the minimax adaptation of overparameterized Bayesian deep neural networks with Gaussian priors . (In progress).

Talks and posters

Teaching