Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster E in Poster Session E: Thursday, August 6, 10:30 am – 12:15 pm, Kimmel Center, Shorin & Rosenthal Rooms

Evidence for Efficient Coding of Naturalistic Images in Visual Short-Term Memory

Kristin M. Dang¹, Manoela Teleginski Ferraz¹, Daniel L. Carstensen², Serra E. Favila², Steven M Frankland¹; ¹Dartmouth College, ²Brown University

Presenter: Kristin M. Dang

Given that both biological and artificial systems must operate with limited past experience and limited processing time, it’s critical they be able to effectively generalize pre-acquired knowledge to new cases. Shepard (1987) famously proposed an overarching principle: the probability of generalizing any property from one stimulus to another decays as a monotonic, approximately exponential function of the distance between the two in a psychological space. Recent work has shown (1) this signature holds for naturalistic images (Marjieh et al., 2024) and (2) can be explained theoretically as a natural consequence of efficient coding (Sims, 2018). Here, we integrate and extend these two lines of work, empirically evaluating whether the same principle holds for human Visual Short-Term Memory (VSTM)– a well-studied capacity-limited communication channel. Specifically, we ask: is the confusability of a target and lure image in VSTM a function of their distance in an independently defined psychological space? We find that the probability of collapsing two naturalistic stimuli in memory is well-modeled as an exponentially decaying function of that psychological distance, providing evidence for the basic principle of efficient coding of naturalistic stimuli in VSTM.

Topic Area: Memory, Learning & Knowledge Structures