Abstract
In spite of the rich, multifaceted nature of our visual memories, research has typically been limited in its ability to assess the neural correlates of memory content, given the short time scales and artificial image sets. However, in an age of social media where people are constantly documenting their lives, meaningful visual memory stimuli are being recorded daily. Here, we leveraged 9,266 daily memory videos recorded by users of the app “One Second Everyday” to revisit key questions on representations of visual memory in the brain. During a 3T fMRI scan, participants (N=23) watched 300 of their memory videos, interwoven with 300 videos from a randomly paired participant. Participant pairs saw the same stream of videos, so that they experienced identical perceptual content but different memory content. Participants also labeled their videos for a wide range of features, including time, location, emotion, strength, and the presence of familiar people and places. We found that these features were highly interconnected: the most memorable events were recent, positive events with new people in distant new locations. Such results might suggest that prior findings of memory age and distance coding in the hippocampus could be partially attributed to the strength and emotion of those memories. Further, computational vision metrics (deep learning image classification, optical flow) could successfully predict a participant’s memory strength for a video, highlighting the need to consider the impact of a memory’s visual features on the ability to retrieve it. Importantly, armed with this rich set of visual memories, we reveal a topography of distinct voxel clusters representing memory time, strength, people familiarity, and place familiarity in the medial parietal cortex. These results highlight a new region of interest for studies of memory content representations and memory consolidation, while also emphasizing the need for more naturalistic study of visual memory.