Recent statistics shift object representations in parahippocampal cortex

Poster Session C - Sunday, April 14, 2024, 5:00 – 7:00 pm EDT, Sheraton Hall ABC
Also presenting in Data Blitz Session 4 - Saturday, April 13, 2024, 1:00 – 2:30 pm EDT, Sheraton Hall EF.

Sarah Solomon¹ (sarahsol@sas.upenn.edu), Kendrick Kay², Anna Schapiro¹; ¹University of Pennsylvania, ²University of Minnesota

There is a stability vs. plasticity trade-off within our semantic system. Our representations need to be stable enough to support our general knowledge of the world but flexible enough to incorporate new information as our environment changes. How does the human brain manage this trade-off? We analyzed the Natural Scenes Dataset in which eight participants viewed thousands of scenes across 30 fMRI sessions. An encoding model learned voxel-wise responses to each of 80 object categories that appeared in the scenes (e.g., giraffe, umbrella) and estimated multivoxel patterns for these objects within each session. We found that multivoxel pattern similarities matched the semantic similarities of a word embedding model across areas in the medial temporal lobe (MTL) and high order visual areas. Within the MTL areas containing semantic representations, only in parahippocampal cortex (PHC) were these representations sensitive to recent co-occurrence statistics in the scenes: the ways that objects appeared together in the first half of a session influenced neural pattern similarity in the second half of the session. Greater mismatch between recent and long-term statistics predicted more change in PHC. We also saw evidence of representational drift in PHC at a longer timescale, across sessions. These results demonstrate that while some regions of the brain encode stable visual object semantics, PHC exhibits higher plasticity, with semantic representations constantly tweaked by the statistics of the recent environment.

Topic Area: LONG-TERM MEMORY: Semantic