Abstract
Which images are better remembered than others has been shown to be highly consistent across adults from different backgrounds, owing to an intrinsic memorability to images. While a number of studies have focused on image memorability among adults, no studies have examined how early children have gained sensitivity to these intrinsic features and determined at what point they start to show adult-like memory patterns to images. This study compared children’s memory of scene images to adults’ memory of the same images modeled by a deep neural network and tested behaviorally. We extracted data on children’s memory from Saragosa-Harris et al. (2021), where 137 children aged three to five years old first encoded animal-scene image pairs and were then cued with animals and asked to indicate the associated scenes from memory after a five-min, 24-h, or one-week delay. Adult memory was predicted by computing memorability scores of the same images using ResMem, a pre-trained deep neural network for predicting memorability (Needell & Bainbridge, 2021). Adults’ memory data was also acquired behaviorally on Amazon Mechanical Turk using a continuous recognition task. Results showed that the memory of scene images among five-year-old children showed significant correlations with the memory predictions by ResMem. ResMem scores were also most predictive of the memories of children across age groups after a one-week delay. These results suggest that adult-like patterns of scene image memories emerge in early childhood by the age of five. Further, these similarities to adults manifest most when memory is tested after a long, one-week delay.