TY - JOUR T1 - Dorsal anterior cingulate-midbrain ensemble as a reinforcement meta-learner JF - bioRxiv DO - 10.1101/130195 SP - 130195 AU - Massimo Silvetti AU - Eliana Vassena AU - Tom Verguts Y1 - 2017/01/01 UR - http://biorxiv.org/content/early/2017/05/23/130195.abstract N2 - The dorsal anterior cingulate cortex (dACC) is central in higher-order cognition and in the pathogenesis of several mental disorders. Reinforcement Learning (RL), Bayesian decision-making, and cognitive control are currently the three main theoretical frameworks attempting to capture the elusive computational nature of this brain area. Although theoretical effort to explain the dACC functions is intense, no single theoretical framework managed so far to account for the myriad of relevant experimental data. Here we propose that dACC plays, in concert with midbrain catecholamine nuclei, the role of a reinforcement meta-learner. This cortical-subcortical system not only can learn and make decisions based on RL principles, but it can also learn to control the learning process itself, for both its own circuits and for other brain areas. We show that a neural model implementing this theory, the Reinforcement Meta-Learner (RML), can account for an unprecedented number of experimental findings include effort exertion, higher-order conditioning and working memory. The RML performs meta-learning by means of approximate Bayesian inference; moreover, it respects several neuro-functional, neuro-anatomical, and behavioral constraints, providing a perspective that assimilates the other theoretical proposals in a single computational framework. ER -