Modeling cognition through adaptive neural synchronization: a multimodal framework using EEG, fMRI, and reinforcement...

Area of Science:

• Computational Neuroscience
• Cognitive Modeling
• Neuroscience

Background:

• Understanding thinking as a neural phenomenon is a key challenge in neuroscience and computational modeling.
• Existing models often lack biological grounding or fail to integrate multiple aspects of brain function.

Purpose of the Study:

• To present a biologically grounded computational framework for simulating adaptive decision-making across cognitive states.
• To integrate neuronal synchronization, metabolic energy consumption, and reinforcement learning within a unified model.
• To validate the model using real EEG and fMRI data.

Main Methods:

• Simulated neuronal synchronization using Kuramoto oscillators, constrained by multimodal activity profiles for energy dynamics.
• Employed Q-learning and Deep Q-Network (DQN) agents to modulate inputs for optimal synchrony and minimal energy cost.
• Validated the model by comparing simulated outputs (spectral power, phase synchrony, BOLD activity) with real EEG and fMRI data.

Main Results:

• The DQN agent demonstrated rapid convergence and superior performance over Q-learning, reducing synchronization error by over 40%.
• The model successfully reproduced canonical brain states (focused attention, multitasking, rest), with accurate spectral power and phase synchrony.
• Cross-modal validation showed moderate correlation with real BOLD signals, and high region-specific prediction accuracy in key cortical areas.

Conclusions:

• The developed framework captures electrophysiological and spatial brain activity, respects neuroenergetic constraints, and adaptively regulates brain-like states.
• This biologically constrained model offers a scalable platform for simulating cognition and developing neuroadaptive systems.
• The study provides a novel approach to modeling thinking as a control problem, paving the way for future brain-computer interfaces.