Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster B in Poster Session B: Tuesday, August 4, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Using Deep Neural Networks To Model The Relationship Between Internal Visual Representations And Aesthetic Appeal

Karen A Wahba¹, Edward A Vessel¹; ¹CUNY City College of NY

Presenter: Karen A Wahba

Aesthetic experience arises from the interaction between perceptual representations, cognitive processes, and affective evaluation. However, how internal visual representations give rise to aesthetic appeal remains unclear. In this study, we introduce latent distinctiveness as a computational measure of representational uniqueness and test its relationship to human aesthetic judgments using a pretrained deep neural network. Latent distinctiveness refers to the relative uniqueness of an image within a learned feature space, quantified as the distance between its embedding and those of other images, and capturing how much an image deviates from typical representations in the visual environment. Across multiple layers of a VGG16 model, latent distinctiveness shows a weak relationship with aesthetic ratings, with the strongest effects emerging in intermediate layers. This pattern is observed across external datasets, suggesting that aesthetic preference depends on the placement of images within broader visual environments. These findings provide a computational account linking hierarchical visual representations to aesthetic evaluation.

Topic Area: Computational Models of Vision & Visual Cortex