SUMMARY / ABSTRACT This proposal aims to reveal critical neural mechanisms for intelligent reward-driven learning and decision-making. It is well established that animals, including humans, use internal models of the world to guide their behavior. Model-based computations are especially vital in complex environments, where animals often need to plan a sequence of choices leading to later rewards. Furthermore, after receiving a reward animals update their reward predictions (“values”) – both for earlier choices they made, and for alternative ways of reaching that reward. These adaptive behaviors rely on combining simulations (of potential paths) with evaluations (of whether they are worthwhile). The hippocampus is believed to play a critical role in model-based decision-making, because it can generate coherent “non-local” representations of distant places. These take at least two distinct forms. First, as rats run through an environment hippocampal place cells generally represent current location, but during late phases of the theta rhythm this place code can sweep forward to potential future locations. It was previously found that these “theta sequences” can rapidly flicker between available choices, on alternating theta cycles – highly suggestive of a role in evaluating options. Second, during immobility the hippocampus can produce sharp-wave-ripple (SWR) events, that often encode highly-compressed sequences of places. These occur much more often shortly after rats receive rewards – highly suggestive of a role in updating values. However, whether hippocampal non-local activity patterns are actually used for these purposes is unknown, largely because whether they access values is unknown. There is intriguing evidence that SWRs evoke activity changes in the network of brain areas vital for value-guided decisions (including frontal cortex and ventral striatum). But whether hippocampal non-local activity is actually associated with value retrieval (during theta sequences) and/or value updating (during SWRs) is unknown. This proposal would rigorously assess the relationships between value-guided decision-making and hippocampal non-local representations. We will employ advanced recording techniques and an innovative foraging task, complemented by sophisticated decoding algorithms and precisely-timed manipulations. We will test two specific hypotheses: that post-reward SWRs are used to update values of distant locations (Aim 1), and that theta sequences are used for online, value-guided decisions (Aim 2). These Aims engage the full combined expertise of our team of experimentalists and theoreticians, including large-scale recordings across multiple regions, statistical methods to identify non-local representations, and reinforcement learning algorithms to estimate decision variables from behavior. These studies have the potential to transform our understanding of how our brains explore internal models to guide learning and decision-making.