[2L4-GS-1] Fundamental AI, theory, algorithm:

[2L4-GS-1-03] Neural computations underlying food preferences

Modeling with deep learning

〇Yuhi Aoki¹, Asako Toyama¹, Yuichi Yamashita², Shinsuke Suzuki¹ (1. Hitotsubashi University, 2. National Center of Neurology and Psychiatry)

Keywords:Computational neuroscience, Cognitive science, Neural network, Deep learning, Food preference

Our daily dietary choices are guided by the subjective values we assign to foods. Yet, little is known about how the subjective values are constructed in the brain. The present study aims to elucidate the neurocomputational processes underlying the food valuation using a deep convolutional neural network model (DCNN). Notably, DCNNs perform at human-level accuracy in object recognition tasks by processing visual information in a manner similar to the brain, making them a suitable computational model for food valuation based on visual inputs. By analyzing rating data for 896 food images provided by 200 participants, we demonstrated that DCNNs can significantly predict the subjective values of foods based on their visual features. Furthermore, an examination of neural activity patterns across the layers of the DCNN revealed that higher-order attributes such as subjective value, healthiness, and tastiness are represented in the later layers, while low-level visual information (i.e., color) is consistently encoded across both early and late layers. These findings suggest that low-level visual information plays a critical role in the entire process of subjective value computation in the brain. Future research will compare DCNN predictions with neuroimaging data to deepen our understanding of the neural computations involved in food valuation.