Theoretical foundations and algorithmic tools for high-dimensional inference: effective model inference beyond maximum likelihood and collective learning
My research project will be centered around the broad field of high-dimensional inference, exploring both its theoretical foundations in statistical physics and disordered systems theory, and applications across machine learning and complex systems. The first and main branch of the project will focus on data-driven modeling, through interpretable generative models in machine learning and statistical physics, with an emphasis on energy-based models (EBMs) and applications to structural inference in computational biology and neuroscience. The main goal is to set-up advanced pair-wise inverse modeling from data with sparsity constraints beyond maximum-likelihood estimation, and exploiting Restricted Boltzmann machines (RBMs) for the effective inference of high-order interactions from data. The second branch instead aims at analyzing planted high-dimensional inference problems in a "collective" learning set-up, where many units (or "students") cooperate to solve a common inference task with fewer data or, alternatively, more efficiently for a given learning algorithm, with applications in the context of federated learning.
This IFISC Seminar will be broadcasted in the following zoom link: https://us06web.zoom.us/j/89027654460?pwd=Wg9TYMPqqP2ipfj2JVvEagmzaTw29c.1
Coffee and cookies will be served 15 minutes before the start of the seminar
Detalls de contacte:
Raúl Toral Contact form