Giacomo Meanti

G. Meanti portrait photograph

I am a PostDoc at INRIA Grenoble in the Thoth team, supervised by Julien Mairal.

I completed my PhD at the University of Genova, supervised by Lorenzo Rosasco.

My research interests include

  • Inverse problems for imaging. I'm interested in novel approaches to solving inverse problems. Recently I've been studying how to deal with the case when the forward model is unknown, and only unpaired datasets are available.
  • Developing efficient algorithms for shallow learning. I developed the Falkon library for scaling up kernel ridge regression, as well as the K-Planes algorithm for representing neural radiance fields with shallow neural networks.
  • Developing kernel methods and shallow learning algorithms for scientific applications. For example dynamical systems, wind forecasting, molecular dynamics.

You can contact me by e-mail at giacomo [dot] meanti [at] gmail [dot] com

Updates

  • 01/05/25DDM4IP was accepted at ICCV 2025. Solving blind inverse problems from completely unpaired data! Code on github.

  • 01/12/24Check out our workshop paper on finetuning foundation models for molecular dynamics using kernels

  • 01/02/24 Joined the Thoth team in Grenoble for a postdoc with Julien Mairal! I'll be working on diffusion models and imaging inverse problems.

  • 11/12/23 At NeurIPS to present our paper on efficient Koopman operator learning

  • 21/06/23 At CVPR to present K-Planes: an efficient NeRF algorithm in arbitrary dimensions.

  • 07/06/23 Our paper on efficient Koopman operator learning with the Nyström approximation, with applications to the study of dynamical systems is available on ArXiv!

  • 30/05/23 I completed my PhD defense! You can check out the slides or the full thesis "Large Scale Kernel Methods for Fun and Profit".