Abstract
Humans and now computers can derive subjective valuations from sensory events although such transformation process is largely a black box. In this study, we elucidate unknown neural mechanisms by comparing representations of humans and convolutional neural networks (CNNs). We optimized CNNs to predict aesthetic valuations of paintings and examined the relationship between the CNN representations and brain activity by using multivoxel pattern analysis. The activity in the primary visual cortex was similar to computations in shallow CNN layers, while that in the higher association cortex was similar to computations in deeper layers, being consistent with the principal gradient that connects unimodal to transmodal brain regions. As a result, representations of the hidden layers of CNNs can be understood and visualized by the correspondence with brain activity. These relations can provide parallels between artificial intelligence and neuroscience.
Competing Interest Statement
The authors have declared no competing interest.
