Francesca Mignacco
Ph.D. student @ IPhT, Université Paris-Saclay
About
Hi! My name is Francesca. I am a Ph.D. student at the Institute of Theoretical Physics, Paris-Saclay University. My thesis is supervised by Lenka Zdeborová (EPFL) and Pierfrancesco Urbani (IPhT) and lies at the crossroads of statistical mechanics and machine learning.
I am interested in understanding the principles underlying learning in neural networks. I work on the high-dimensional dynamics of stochastic optimisation algorithms, investigating the reasons behind their good generalisation properties.
Contact: francesca.mignacco@ipht.fr
CV [PDF]
Publications
See also my Google Scholar page!
Learning curves for the multi-class teacher-student perceptron [PDF]
Elisabetta Cornacchia, Francesca Mignacco, Rodrigo Veiga, Cédric Gerbelot, Bruno Loureiro, Lenka Zdeborová
arXiv preprint: arXiv:2203.12094
The effective noise of Stochastic Gradient Descent [PDF]
Francesca Mignacco, Pierfrancesco Urbani
arXiv preprint arXiv:2112.10852 (2021).
Stochasticity helps to navigate rough landscapes: comparing gradient-descent-based algorithms in the phase retrieval problem [PDF]
Francesca Mignacco, Pierfrancesco Urbani, Lenka Zdeborová
Machine Learning: Science and Technology (2021).
Dynamical mean-field theory for stochastic gradient descent in Gaussian mixture classification [PDF]
Francesca Mignacco, Florent Krzakala, Pierfrancesco Urbani, Lenka Zdeborová
Thirty-fourth Conference on Neural Information Processing Systems (NeurIPS 2020).
The role of regularization in classication of high-dimensional noisy Gaussian mixture [PDF]
Francesca Mignacco, Florent Krzakala, Yue M. Lu, Pierfrancesco Urbani, Lenka Zdeborová
Thirty-seventh International Conference on Machine Learning (ICML 2020).