Computational modelling of hippocampal sensitivity to expectation violation

Abstract

Pattern separation and completion are fundamental hippocampal computations supporting memory encoding and retrieval. However, despite extensive exploration of these processes, it remains unclear whether they are modulated by top-down processes. We used a neural network model to examine how unexpected information is represented by the hippocampus. During training the network learned a contingency between a cue and a category, which the target object belongs to. At test, we presented the network with congruous and incongruous cues, as well as perceptually similar foils. We used representational similarity analysis to examine how the top-down expectation modulation interacts with bottom-up perceptual input, in each layer. All subfields showed an interaction between the two, with DG and CA3 being more sensitive to expectation violation than CA1. A further multivariate analysis revealed that representational differences between expected and unexpected inputs were prominent for moderate to high levels of perceptual overlap in DG/CA3. This effect diminished when inputs from DG and CA3 into CA1 were lesioned. Overall, our findings suggest pattern separation in DG and CA3 underlies the effect that violation of expectation exerts on memory.