Examining how Pattern Matching for Multiple Distributions is Impacted by Time

Abstract

It has long been suggested that memories continue to be analyzed after initial encoding. This idea sets the basis for the Complementary Learning Systems theory that suggests the hippocampus and cortex have two separate roles in consolidating memory. This theory proposes that the hippocampus rapidly memorizes individual instances and then replays these instances to the cortex where over time commonalities can be distinguished and extracted across memories. However, recent research has also indicated the hippocampus may be able to rapidly extract regularities. Therefore, this thesis set forth to examine how representations for two spatial distributions, when exposed to them during the same single-day 30-minute training period, compared depending on time allotted for extracting regularities. To test this, twenty human subjects explored a virtual circular maze. During training, participants went through 30 trials where they had to locate a platform taken randomly from one of two spatial distributions (15 trials for each distribution). Following training participants returned 1 day or 30 days later to complete the testing portion of the study. During testing the participants explored the same virtual world but this time in the absence of platforms to see how well they remembered the two distributions. To test how well they had consolidated the information from the training trials the participants’ test traversal patterns were compared to the true underlying distributions using Kullback-Leibler divergence. Results showed that individuals in the 1- day test group had significantly better representations for both individual spatial distributions compared to the 30-day test group. However, both groups had similar representations when looking at the combined overall distribution of all the platforms, irrespective of their initial grouping. These results suggest that multiple memories associated with each different spatial distribution do not continue to be processed separately for a long time after initial encoding. While individual representations are maintained for each spatial distribution for at least 24 hours this thesis provides some evidence that over time the multiple memories from both distributions get consolidated together resulting in a combined representation.