eLife Assessment
This study makes a valuable contribution to the understanding of meta-learning and its neural mechanisms by distinguishing two timescales of learning rate adaptation: rapid, within-block reductions and slower, location-specific, meta-learned adjustments. Behavioural data and computational modelling provide convincing evidence that individuals adjust learning rates both rapidly in response to uncertainty and more gradually through meta-learning of environmental statistics. Neuroimaging results indicate that meta-learned learning rates are represented in orbitofrontal cortex, and that prediction errors are encoded across a distributed network including the ventral striatum, where they are modulated by expectations about error magnitude. The manuscript is timely and clearly written and opens the door to future work on how these signals contribute to adaptive behaviour.