UC Riverside

Human memories do not always precisely correspond to exceedingly rich contents in the external environment. This variability of internal representations, especially in working memory (WM) – a system that maintains a small amount of information over a short time period at the service of other mental activities, sets an important functional limit in human cognition. However, neurocognitive mechanisms underlying this WM precision bottleneck remain unclear. One class of theories attributes WM precision to noisy sustained neural activity that supports WM retention (i.e., neural noise hypothesis). Another class of theories maintains that WM retention and precision are supported by independent neural mechanisms. Specifically, WM precision and its manifestation as a certain level of noise in sustained neural activity may be supported by pattern separation, a computation potentially implemented in the hippocampus to orthogonalize similar memories into non-overlapping representations. This pattern separation hypothesis is preliminarily supported by 3 lines of evidence in 4 experiments of the current dissertation. First, in Experiment 1, observers with better pattern separation performance in a behavioral task tend to have higher precision in both WM and long-term memory (LTM), in contrast to a lack of significant association between pattern separation behavioral performance and the probability of successful remembering in either WM or LTM. Second, using functional Magnetic Resonance Imaging (fMRI), Experiment 2 shows that the hippocampus, along with several other regions in a distributed neural network, increases activity as the task demand on WM precision increases. This hippocampal sensitivity to WM precision task demand seems to be primarily driven by the DG/CA3 subfield – where pattern separation most likely occurs – in Experiment 3 using high-resolution fMRI. Third, Experiment 4 further demonstrates that the hippocampal DG/CA3, during WM delay period, retains decodable item-specific information, which further predicts activity in the visual cortices, potentially linking pattern separation to sensory recruitment for precise visual WM representations. Overall, these findings support a novel hippocampal pattern separation mechanism for WM precision, which is central to the ongoing debate on the nature of WM storage limitations. Articulating this potential mechanism may provide a better understanding of compromised mental clarity, manifested as reduced memory precision, in clinical populations such as schizophrenia and pathological aging, etc.