A main focus in the past years has been on the analysis of fMRI data to explore the brain coordination at rest or when humans perform a task. A whole-brain dynamic model is used to estimate the directional interactions between brain regions. These interactions are denoted by whole-brain effective connectivity (EC), moving beyond a simple phenomenological description of their observed correlations (functional connectivity). The model implies simpler assumptions than dynamic causal modeling.
The modifications of the EC pattern at the network level hint at the selection of specific pathways for, e.g., sensory integration.
I also plan to extend the existing framework to interpret EEG and MEG data.
During my MSCA project, I developed a network-specific analysis for multiunit activity recorded in monkeys in collaboration with Prof Alex Thiele (Newcastle University). The goal was to compare the effective connectivity estimated from observed network activity to a-priori knowledge about anatomical connections.
With Adrià Tauste Campo from BarcelonaBeta, we've developed a non-parametric method for connectivity estimation from observed network activity, based on the multivariate autoregressive (MVAR) process, which can be seen as an alternative to Granger causality analysis.
I am interested in learning representations in spiking networks, for example using models of synaptic plasticity like STDP. Neuronal representations are an alternative terminology to neuronal coding and refer to transformations of the spiking statistics. For example, sensory inputs correspond to stimulus-dependent spiking statistics; for spiking rates (i.e., means) in the visual system for gradings with various orientations, they correspond to tuning curves.
Mathematical modeling formalizes the input-output mapping implemented by the network dynamics, which is shaped by learning. Recurrently-connected networks involve non-linearities due to the global feedback. This induces an intricate relationship between network dynamics, as well as learning.