Vision-to-value transformations in artificial neural networks and human brain

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.

The Role of the Ventromedial Prefrontal Cortex in Preferential Decisions for Own- and Other-Age Faces

“…how subjective value was formed through bottom-up processing (Iigaya et al, 2020;Pham et al, 2021) indicates that the extended BVS might serve as a gateway for domain-specific value computation or attribute value computation (Lim et al, 2013). A natural interpretation of the higher pleasantness-rating scores for young faces found in both age groups is that young faces are more rewarding than older faces, regardless of participants' age.…”

“…The researchers found a functional coupling between the ventral striatum and the auditory cortex, which represented subjective values of musical excerpts. A recent understanding of how subjective value was formed through bottom-up processing ( Iigaya et al, 2020 ; Pham et al, 2021 ) indicates that the extended BVS might serve as a gateway for domain-specific value computation or attribute value computation ( Lim et al, 2013 ).…”

Applications of artificial intelligence in dementia

“…A CNN works by repeatedly passing various filters over images, which strongly resembles the neural visual information processing of the brain in the primary visual cortex (V1). [7][8][9] This is a compelling example of using computers to model the brain and gain a deeper understanding of its functions. Of the two visual pathways (dorsal and ventral) (Fig.…”

Constructing value signals for food rewards: determinants and the integration

“…Remarkably, the hierarchical architecture (Figure 2a) resembles the proposed mechanism for the valuation of the visual arts. Recent studies [32,33] have suggested that hierarchical information processing underlies the valuation of visual arts by correlating a deep convolutional neural network (DCNN) model with behavioural and neuroimaging data. DCNN is a machine learning algorithm that performs visual object recognition tasks, as accurate as humans [34].…”

“…Furthermore, they found a significant correspondence between the earlier (deeper) layers in the artificial neural network and visual cortical (parietal-frontal) regions in the human brain. The similarity between the valuations of food items (Figure 2) and visual arts [32,33] implies that the hierarchical processing of information is a ubiquitous structure for constructing value signals independent of the type of object [16]. A fruitful avenue for future research would be to elucidate the hierarchical mechanism for food value computations by leveraging DCNN models.…”