Investigate Attractor Formation in Working Memory through Frequency Manipulation of Natural Images

Abstract

Working memory (WM) plays a critical role in maintaining and manipulating visual information in the absence of sensory input. While attractor dynamics, where memory representations stabilize around specific locations in memory space, have been observed in simpler visual domains like color or orientation, it remains unclear whether similar mechanisms apply to complex stimuli such as natural faces. The present study aims to explore whether repeated exposure to specific faces can induce attractor dynamics in working memory for natural images. Using a continuous report paradigm with VAE-generated face stimuli, participants were shown target faces at varying frequencies and later asked to recall the target face from a set of images. Although some participants showed memory biases and attractor-like structures near the target face, these effects were not robust across the group. A statistical learning phase, where participants were passively exposed to a target face, did not induce a consistent shift in memory responses. These results suggest that while attractor dynamics may generalize to natural stimuli, they are resistant to passive statistical learning manipulations, highlighting the need for stronger or behaviorally relevant interventions to modulate attractor dynamics. Future research should investigate the impact of more active learning strategies, cognitive load, and long-term memory tasks on attractor stability and flexibility.