Beyond Motor Noise: Considering Other Causes of Impaired Reinforcement Learning in Cerebellar Patients.

Abstract

Motor and reinforcement learning have been classically linked to functionally independent brain networks centered on the cerebellum and the basal ganglia, respectively. In a recent study published in eNeuro , Therrien et al. (2018) showed that increasing motor noise in healthy subjects disrupts reinforcement learning. However, this impairment remained well below that detected in cerebellar patients even when motor noise in healthy subjects was adjusted to match that observed in the patients. This suggests that impaired reinforcement learning following cerebellar damage cannot be solely accounted for by altered motor noise in these patients. Based on recent anatomic and functional evidence, we argue that the cerebellum may directly contribute to reinforcement learning, consistent with its tight connections with the basal ganglia. The ability to adapt to changes occurring in the environment is a fundamental feature of human behavior, which relies on both sensory and reward feedbacks. On the one hand, the role of sensory feedback has been largely considered by studying how motor commands adapt to visual perturbations (e.g., a visuomotor rotation), a process called error-based learning (Shadmehr et al., 2010; Wolpert et al., 2011; Kim et al., 2018; Roemmich and Bastian, 2018). This type of motor learning involves the computation of sensory prediction errors (SPEs), namely, the difference between predicted and actual sensory outcome (Tseng et al., 2007; Schlerf and Ivry, 2012; Shadmehr, 2017, 2018). On the other hand, the role of reward feedback has been mostly investigated in tasks that require learning what action to select or not, by updating reward predictions based on previous experience, a process named reinforcement learning (Lee et al., 2012; Derosiere et al., 2017a,b; Gershman and Daw, 2017; O’Doherty et al., 2017). A central …