Cognitive control is a set of different mechanisms that our brain has to take our plans and abstract goals and translate them into actions. Abstract goal representations in prefrontal cortex (PFC) are thought to provide top-down influences over premotor and motor cortices to guide action selection. Neuroimaging and lesion studies have shown that PFC is organized hierarchically along a rostral-to-caudal functional gradient, from higher-order abstract rules to lower-order concrete action representations. Recent studies suggest that this functional hierarchy is mediated by local regions of the striatum via fronto-striatal loops, characterized by a rostral-to-caudal organization. Parietal regions have similar, order-specific connectivity with the striatum, but along the caudal-to-rostral axis. Together these previous findings suggest that separable, distributed association networks are involved in hierarchical cognitive control.

In collaboration with Justin Riddle, I am using a multimodal approach to test the hypothesis of a multiplexed model of hierarchical control, which operates through functional interactions at different frequencies for the control of stimulus-action associations (sensorimotor) and contextual control (stimulus-action associations dependence on context). We show that top-down delta oscillations mediate the frontal-parietal synchronization between LPFC and IPL for contextual control. Theta oscillations serve as control signals for integrating information relative to sensorimotor control, in more caudal and midline regions of the frontal cortex. Each network is supported by connections with specific portions of the dorsal striatum. Overall, our findings support a model of dynamic cognitive control processing through multiplexed synchronization mechanisms.

Working memory (WM) is our ability to temporarily hold in mind and manipulate information. WM is acknowledged as having limited capacity and precision, as well as being characterized by specific biases and errors. In particular, when subjects are asked to maintain multiple items in WM and to report a feature of one item (e.g. spatial location or color), the so-called swap errors occur when an inaccurate response to the target item is accurate relative to a non-target item. These errors reflect the failure to maintain the correct binding between the individual features that define each item (location/color).

A recent biophysical neural network model proposes that item features in WM are bound through low-frequency network synchronization, and swap errors can result from a disruption of such synchronization. I test the neurophysiological prediction of this model by using MEG data collected from human subjects, in a task designed to induce binding errors. The goal of this project, which is carried out in collaboration with the Sreenivasan Lab (NYU Abu Dhabi), is to characterize the neural mechanisms supporting feature-binding in WM.

Visual selective attention allows us to optimize the use of our limited cognitive resources, by preferentially processing relevant sensory information over irrelevant information. This fundamental brain mechanism is mediated by a distributed large-scale network of brain regions, where fronto-parietal areas in particular are thought to play a role of control. The dynamics of attentional control signals remain an area of active investigation.

To characterize the neural dynamics of visual selective attention, I use EEG source-reconstruction techniques and assess the attention-induced local and connectivity changes, in a network of fMRI-defined regions of interest. A key finding of this work is that selective attention is mediated by a fast reorganization of frequency-specific inter-areal interactions. Furthermore, the posterior parietal cortex (PPC) acts as a gatekeeper for attentional processing, by selectively enabling or down-regulating activity in downstream areas through top-down gating influences. These influences change quickly during the selection of attended visual features, with paths and dynamics that depend on the functional specialization of regions along the visual stream.

Brain cognitive networks rely on frequency-specific inter-areal interactions. These networks can flexibly and rapidly reorganize, changing interactions properties on subsecond time scales. This ability allows our brain to efficiently respond to environmental changes in goal-directed ways, but it also makes it difficult to investigate these phenomena. In electrophysiological data, inter-areal interactions are typically modeled from statistical regularities between the signals simultaneously recorded from different brain regions; for example, using directed connectivity measures based on spectral Granger–Geweke causality (GGC). Notably, GGC measures can also be estimated dynamically, using time-varying methods that depend on either spectral decomposition (nonparametric) or multivariate autoregressive modeling (parametric) of the recorded signals.

I employ a combination of numerical simulations and benchmark data to systematically assess the performance of existing time-varying connectivity methods, proving their efficacy in correctly characterizing rapidly changing, dynamic interactions. In related works, I highlight strengths and weaknesses of these methods, suggesting best practices for application and highlighting possible interpretability limitations. This research led also to the development of a novel improved parametric algorithm, as well as a toolbox that comprises different nonparametric methods.