Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session A: Tuesday, August 12, 1:30 – 4:30 pm, de Brug & E‑Hall

Testing a Learning Theory of Aesthetic Appeal Using Category Learning and Deep Neural Networks

Edward A Vessel¹, Andrew Frankel, Aubrey Valdez, Aishwarya Gurung, Colin Conwell²; ¹CUNY City College of NY, ²Johns Hopkins University

Presenter: Edward A Vessel

How do people mentally represent visual art, and how do those representations relate to aesthetic value? The learning theory of aesthetic valuation suggests that the aesthetic appeal we feel from engaging with visual objects is an affective signal for learning, and thus depends on how those objects relate to what we know about the visual world. Yet this theory is hard to test, given the difficulty of directly measuring the relevant aspects of an observer's internal perceptual models. We outline a behavioral and modeling paradigm for training observers in a visual artwork training task and, in parallel, tuning deep neural networks (DNNs) to serve as proxies for internal representations. Here we show that the task successfully modulated observer's knowledge and internal representations about a set of artworks, and we explore how architecture and training target affect the ability of DNNs to capture salient aspects of human observers' behavior.

Topic Area: Visual Processing & Computational Vision

Extended Abstract: Full Text PDF