The features underlying the memorability of objects

Abstract

What makes certain images more memorable than others? While much of memory research has focused on participant effects, recent studies using a stimulus-centric perspective have sparked debate on the determinants of memory, including the roles of semantic and visual features and whether the most prototypical or atypical items are best remembered. Prior studies have typically relied on constrained stimulus sets, limiting a generalized view of the features underlying what we remember. Here, we collected more than 1 million memory ratings for a naturalistic dataset of 26,107 object images designed to comprehensively sample concrete objects. We establish a model of object features that is predictive of image memorability and examined whether memorability could be accounted for by the typicality of the objects. We find that semantic features exert a stronger influence than perceptual features on what we remember and that the relationship between memorability and typicality is more complex than a simple positive or negative association alone.

Children develop adult-like visual sensitivity to image memorability by the age of 4.

“…One possible explanation is that the information used by ResMem to predict children’s memory better matches the information remaining in children’s memory after a long delay. Specifically, given the decay of visual information in memory over time observed in adults (Lampinen et al, 2001; Potter, 2012), it is possible that richer semantic features of an image—those shown to be most predictive of memorability (Kramer et al, 2023)—is what persists in children’s memory after a delay. Although the memorability effect remains stable with different delays among adults (Isola et al, 2014; Lin et al, 2021), it is unclear whether it pertains across time among children.…”

“…In addition, the accumulation of language inputs leads to a rapid improvement in young children’s language skills, which may allow children to extract high-level information and encode abstract representations of scene images by 4 years of age. Abstract, semantic information may be the type of features that predominantly drive consistencies in memorability (Kramer et al, 2023; Needell & Bainbridge, 2022).…”

“…In addition, the accumulation of visual experience in the first few years of life could gradually train children’s visual memory to become more adult-like. Memorability may reflect the statistical properties of stimuli in our world (Hovhannisyan et al, 2021), such as co-occurrence statistics (Xie et al, 2020) or across-item similarity (Lukavský & Děchtěrenko, 2017, although see Kramer et al, 2023). Thus, consistencies in memory could develop as children learn these statistical regularities of visual stimuli.…”

Deep neural network decodes aspects of stimulus-intrinsic memorability inaccessible to humans.

“…Lastly, our demonstration of the dissociability for humans' and ResMem's memorability predictions was limited to static images of individuated objects and scenes. Given that a recent study demonstrated memorability for dynamic stimuli such as dance moves (Ongchoco et al, 2023), future studies should also explore whether and how humans and deep neural networks can predict memorability for such dynamic Going beyond, future studies should also explore memorability in other stimulus domains, such as verbal stimuli (Aka et al, 2023;Kramer et al, 2023;Madan, 2021). With the recent development in large language models, more insights could be gained from comparing machine learning and human observers in predicting memorability using verbal and even syntactic materials.…”

“…However, both ResMem and humans are not perfect at predicting the intrinsic memorability of a stimulus. Given that memorability is composed of multiple factors spanning across perceptual and semantic features of a stimulus (Isola et al, 2014; Kramer et al, 2023; Rust & Mehrpour, 2020), it is possible that ResMem and humans utilized dissociable factors to make memorability predictions. To test this possibility, we examined whether ResMem could predict the subjective memorability judgments that were made by humans in Saito et al (2023).…”