Short CV

Maurizio Mattia, Physics by training, received a PhD in Neurophysiology from Sapienza University of Rome, Italy. Afterward, he worked as a postdoctoral researcher at the Istituto Superiore di Sanità (ISS) and previously at the University of Genoa, Italy, focusing on theoretical physics and the computational principles underlying neuronal dynamics.  Since 2002, he has been a Researcher and later a Senior Researcher at the National Center for Radiation Protection and Computational Physics of the ISS in Rome. His research focuses on computational and theoretical neuroscience, neural population dynamics, and modeling of brain activity. In particular, he works on recurrent neural network models, attractor dynamics, stochastic processes in spiking neural populations, and mechanistic modeling of brain states, with applications ranging from motor planning to cortical dynamics and neurophysiological signal analysis. He has authored numerous peer‑reviewed publications, contributed to major European Commission–funded projects, and serves as associate editor for leading journals in computational neuroscience.

Title of the talk

RNN-based digital twins of brain activity for longitudinal trajectory mapping

Abstract

Characterizing how brain function evolves over time at the individual level remains a major challenge in longitudinal neuroimaging. In this talk, I will present results from Di Antonio et al., Advanced Science (2026), where we introduced a computational framework for mapping subject‑specific trajectories from resting‑state fMRI data. Using reservoir computing, our method derives stable linear stochastic recurrent neural networks (RNNs) that generate BOLD time series statistically equivalent to those recorded for each subject. These RNN‑based digital twins are fully characterized by their Laplace spectra, which serve as compact subject fingerprints at a given time point. This representation enables embedding subjects into a topology‑preserving landscape that captures both global population structure and fine‑grained individual progression. I will show how this approach allows us to delineate distinct longitudinal pathways and isolate the mechanistic factors driving trajectory evolution. By situating each participant’s temporal dynamics within a shared functional landscape, we move toward more dynamic, mechanistic, and individualized interpretations of resting‑state fMRI in bioengineering and neuroscience applications.