Contribution of Eye Movement Patterns to Representational Drift
Poster Presentation: Tuesday, May 20, 2025, 8:30 am – 12:30 pm, Pavilion
Session: Eye Movements: Cognition
Schedule of Events | Search Abstracts | Symposia | Talk Sessions | Poster Sessions
Yixin Yuan1, Mikio Aoi1, John Serences1; 1UC San Diego
Representational drift, the phenomenon where neural activity patterns change in a systematic manner when exposed to the same stimulus, has been observed in mouse and human primary visual area (V1). However, given the complexity of movies and natural images, it’s unknown whether the source of this change is intrinsic to neural computation or if animals and humans are covertly attending to a different set of features each time an image is presented. To investigate whether systematic changes in gaze or attention can contribute to representational drift, we ran a longitudinal eye-tracking study with complex visual stimuli repeated over time. 10 healthy adults (8F, 2M) participated in a free-viewing study over 6 sessions that spanned ~2 - 4 weeks. We tracked eye position while participants were presented with naturalistic images that repeated either within, across, or both within and across sessions. Fixation density distribution was estimated for each image to highlight locations that were most commonly foveated within each trial. Based on this density distribution, we summed the normalized fixation likelihoods across all presentations of a given image to form a saliency map. The information accumulation rate for each image was then quantified as the cumulative saliency values from fixation to fixation. This cumulative function can be compared across sessions to investigate differing temporal characteristics of gaze on the same image repeated over time. While there were substantial differences across individuals, when averaged across participants, images presented in earlier sessions had a faster information accumulation rate than those presented in later sessions. This indicates that gaze dynamics change systematically over sessions, suggesting one source of change that may contribute to representational drift. While speculative, these observations suggest that covert top-down cognitive factors such as shifts in attentional focus and eye position may bias neural encoding in a directional manner over time.