Brain Dynamic Functional Connectivity for Artificial General Intelligence

At the ELIXIR Lab, we investigate how the brain’s dynamic functional connectivity can inspire the development of more general and adaptive artificial intelligence systems. Our research bridges neuroscience, machine learning, and embodied cognition to explore how patterns of neural activity evolve over time and how these patterns can inform the architecture of brain-inspired AI.

A central focus of this work is modeling the time-varying interactions between distinct brain regions during cognitive tasks such as memory recall. We process electrophysiological data using source localization techniques to reconstruct 3D brain activity, which is then mapped onto anatomically and functionally defined cortical areas. This enables us to work with structured time-series signals that reflect meaningful regional dynamics in the brain.

To analyze and predict how brain states evolve, we train recurrent neural networks on these signals, capturing temporal dependencies that mirror cognitive flow and perception-action loops in the brain. By interrogating the internal representations of these networks, we infer directed patterns of influence among brain regions—offering a new computational lens through which to study dynamic cognition.

Our research also develops hierarchical models of brain connectivity that reflect the layered, nested structure of neural processing. These models not only help us understand the temporal organization of memory and multisensory integration, but also point toward architectures for artificial general intelligence that emulate the brain’s ability to flexibly coordinate distributed activity.

Ultimately, our work in this area seeks to reverse-engineer the brain’s strategies for learning, prediction, and adaptation—and apply them toward building intelligent systems that are not just task-specific, but capable of general-purpose reasoning and real-world autonomy.

Selected Publications

To come