Neural basis of transferable representations for efficient learning

We learn different rules that can switch depending on context. When the rule depends on context, the context is important to learn to differentiate the rewarded choice in different contexts. While it has previously been shown that the reinstatement of context guides decisions for reward, it is less clear how or whether context-based reactivation guides the efficient learning of reward and context associations. We hypothesize that representations of the context are used to guide learning of associations between context and reward. We test whether this context is retrieved from memory circuits prior to when a choice is made to optimize decisions. In this study, we assess how context representations in the high frequency 80-120 Hz ripple band transform and are retrieved using brain-wide oscillations recorded with intracranial EEG in 14 human participants who completed a rule learning task. We found that (1) context-related high frequency oscillatory patterns are reactivated for context dependent decisions when a reward is received, as indicated by good decoding prediction of context, (2) 4-8 Hz theta activity in medial temporal lobe and frontal cortex accompanies these reactivations, and (3) the dependence of decisions on context influences the speed of evidence accumulation necessary to reach a decision, modelled as a drift diffusion process. These findings suggest that coincident reactivation of context representations and theta activity in memory circuits during rewards for correct choices may influence subsequent decisions.