Brain self-regulation learning in the neurocomputational framework of active inference

Abstract

Neurofeedback (NF), a cutting-edge technique in the realm of brain-computer interfaces (BCI), has proven to be a powerful tool for both scientific exploration and clinical rehabilitation. NF provides individuals with real-time information about their neural processes, enabling them to modulate and regulate their brain activity—a phenomenon known as 'brain self-regulation learning'. While NF holds great promise, it faces an efficiency hurdle. Remarkably, only 50% of participants successfully achieve self-regulation, limiting its clinical adoption. Existing models have struggled to fully elucidate the intricate interplay between reward mechanisms and cognitive functions, without fully succeeding. Herein lies the significance of Active Inference—a theoretical framework that illuminates this complex relationship. To address this gap, we propose using the framework of Active inference to understand the neural processes underlying self-regulation learning. Active inference provides a statistical model of the brain and a combination of computational modeling and neuroimaging techniques. By analyzing real-time functional MRI data and implementing agent-based simulations, we identify that learners exhibit a hierarchical computational anatomy as the neural substrate that supports the internal dynamics of the brain. Our findings underscore the importance of cognitive processes in self-regulation learning and provide insights for optimizing NF protocols.