CCN meeting | CANCELLED Massimo Silvetti (Institute of Cognitive Sciences and Technologies, Italy), invited by Tom Verguts

This meeting has been cancelled.

Meta-Reinforcement Learning as a unifying perspective about the neuro-cognitive mechanisms of curiosity, control, and motivation.

Computational and cognitive neuroscience research on decision-making and cognitive control has so far evidenced the prevalence of three different theoretical approaches in reciprocal competition: expected value of control, surprise and performance monitoring, and, most recently, active inference. Here, we propose a different theoretical framework, based on meta-reinforcement learning, that provides a unified neuro-computational theory for understanding the heterogeneous experimental results from the research domains of cognitive control, effort-based decision-making, foraging behaviour, and information gathering, bridging all the previous main theoretical perspectives and showing a better capability to predict experimental data from different domains. Our theoretical perspective is instantiated in the Reinforcement Meta-Learner (RML), a neuro-computational model proposing that adaptive control of brainstem neuromodulatory activity by the prefrontal cortex influences learning rate, cognitive and physical effort, and motivation, changing the agent–environment interactions (behaviour), which, in turn, modify the brainstem control dynamics, thus closing the meta-learning loop. We finally present novel results about the experimental testing of the RML’s predictions through model-based fMRI analysis of an effort-based decision-making task under stressful and non-stressful contexts.