Memorability: Differences Between Children and Adults

Abstract

Introduction

Research has shown that image memorability – i.e., if a certain image is memorable or forgettable – is consistent among people and is therefore not largely affected by individual differences and past life experiences [1]. Image memorability can be predicted and neural networks are better at doing so than humans – the attributes that make an image memorable seem to be different than what humans expect [2].

Previous research has focused mainly on memorability in healthy adult individuals. This thesis aims to assess the memorability of images in children and to find out if any differences between adults and children occur. With this, we want to understand if the memorability of images changes or is consistent in development.

Methods & Expected Results

150 healthy adults aged 20 to 25 years and 150 children aged 6 to 7 years will be included in the thesis. All participants will be Slovenian. We will conduct a 40-minutes long experiment. The participants will look at different pictures and will have to signalize when they recognize that a picture is shown again. We will get a rate of image memorability for a specific picture separately for adults and children and we will compare these rates with predicted memorability scores form an already existing neural network [2].

We expect to find differences in image memorability between children and adults and that the existing neural network that was only trained on adults will thus successfully predict the image memorability in adults, but not in children. Because no such research was conducted before, we will also try to evaluate which kind of images are more memorable among children and which among adults and we will retrain the neural network separately for the two groups in order to get a neural network that can also accurately predict memorability scores in children.

Limitations & Conclusion

One limitation is that no similar experiment was done before and we can therefore not accurately predict if and what differences between subjects will occur. Another limitation is that the differences observed could stem from generation and not development differences. A longitudinal study or multiple experiments with different generations would be better in this case.

Alongside better understanding of the development of memorability, this thesis also aims to find out which images are memorable in children in order to incorporate such images in schools to improve the memorability of the learned subjects.

References

[1] P. Isola, J. Xiao, A. Torralba and A. Oliva, “Understanding the Intrinsic Memorability of Images,” Journal of Vision, vol. 12, no. 9, August 2012.

[2] C. D. Needell and W. A. Bainbridge, “Embracing New Techniques in Deep Learning for Estimating Image Memorability,” Journal of Vision, vol. 21, no. 9, e1921, September 2021.