Maître de Conférences

Laboratoire de Probabilités, Statistique et Modélisation (LPSM)

Université Paris Diderot

e-mail : giulini @ math.univ-paris-diderot.fr

bureau : 5006 (bat. Sophie Germain)

2017 - pres.  Maitre de Conference at Université Paris Diderot

2016 - 2017  ATER at Université Paris Diderot

2015 - 2016  post-doc at INRIA Saclay, team DataShape


2012 - 2015  Ph.D. at DMA, Ecole Normale Superieure, Paris
    granted by DIM Rdm Idf and Region Ile-de-France

    PhD defense: 24 September 2015
    Generalization bounds for random samples in Hilbert spaces
    supervisor: Olivier Catoni


2007 - 2012  Università degli Studi di Genova

    20 June 2012, M.Sc Degree in Mathematics
    Compressive sensing and sparse recovery

    supervisor: Ernesto De Vito

    21 July 2010, B.Sc. Degree in Mathematics
    Spazi di Hilbert a nucleo riproducente: Teorema di Mercer
    supervisors: Ernesto De Vito, Emanuela Sasso

my main research field is statistical learning. In particular, I am working on

• high-dimensional data analysis
  

• PAC-Bayesian learning
  

• spectral clustering

PhD thesis Generalization bounds for random samples in Hilbert spaces

preprints

• Kernel Spectral Clustering; I. Giulini; (2016), http://arxiv.org/pdf/1606.06519v1.pdf  

• Dimension-free PAC-Bayesian bounds for matrices, vectors, and linear least squares regression; O. Catoni, I. Giulini; (2017), https://arxiv.org/pdf/1712.02747.pdf

publications

2017 

• Dimension free PAC-Bayesian bounds for the estimation of the mean of a random vector; O. Catoni, I. Giulini; NIPS-2017 Workshop (Almost) 50 Shades of Bayesian Learning: PAC-Bayesian trends and insights

• Robust PCA and pairs of projections in a Hilbert space; I. Giulini; Electronic Journal of Statistics, 11(2):3903-3926, 2017

• Robust dimension-free Gram operator estimates; I. Giulini; Bernoulli Journal, 24(4B):3864-3923, 2018 https://arxiv.org/pdf/1511.06259.pdf

2016

• Data driven estimation of Laplace-Beltrami operator; F. Chazal, I. Giulini, B. Michel; NIPS-2016,
• long version https://hal.inria.fr/hal-01387021