Deep learning of individual aesthetics
- PDF / 5,532,093 Bytes
- 15 Pages / 595.276 x 790.866 pts Page_size
- 16 Downloads / 163 Views
(0123456789().,-volV)(0123456789(). ,- volV)
S. I : NEURAL NETWORKS IN ART, SOUND AND DESIGN
Deep learning of individual aesthetics Jon McCormack1
•
Andy Lomas2
Received: 22 June 2020 / Accepted: 18 September 2020 Ó Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract Accurate evaluation of human aesthetic preferences represents a major challenge for creative evolutionary and generative systems research. Prior work has tended to focus on feature measures of the artefact, such as symmetry, complexity and coherence. However, research models from psychology suggest that human aesthetic experiences encapsulate factors beyond the artefact, making accurate computational models very difficult to design. The interactive genetic algorithm circumvents the problem through human-in-the-loop, subjective evaluation of aesthetics, but is limited due to user fatigue and small population sizes. In this paper, we look at how recent advances in deep learning can assist in automating personal aesthetic judgement. Using a leading artist’s computer art dataset, we investigate the relationship between image measures, such as complexity, and human aesthetic evaluation. We use dimension reduction methods to visualise both genotype and phenotype space in order to support the exploration of new territory in a generative system. Convolutional neural networks trained on the artist’s prior aesthetic evaluations are used to suggest new possibilities similar or between known highquality genotype-phenotype mappings. We integrate this classification and discovery system into a software tool for evolving complex generative art and design. Keywords Evolutionary art Aesthetics Aesthetic measure Convolutional neural networks Dimension reduction Morphogenesis
1 Introduction Computational evolutionary methods can support human artists and designers in exploring the aesthetic possibilities of complex generative systems [3, 4, 31]. However, the majority of evolutionary algorithms used for art and design rely on being able to compute a fitness measure of phenotype aesthetics. Devising formal aesthetic measures is a long-standing, but generally illusive quest in evolutionary computing and psychology research [5, 16, 18]. As a way of circumventing the formalisation of an aesthetic measure, the interactive genetic algorithm (IGA) has long been used by artists and researchers since first being devised by Dawkins in the mid-1980s & Jon McCormack [email protected] Andy Lomas [email protected] 1
SensiLab, Monash University, Caulfield East, Australia
2
Goldsmiths, University of London, London, UK
[10, 30, 32, 35, 37, 40, 41]. A key advantage of the IGA is that it puts a ‘‘human in the [evolutionary] loop’’, substituting formalised fitness evaluation for human judgement. To evolve a visual form, the user simultaneously assesses or compares a small population (typically around 16–25 individuals) from a single parent (if offspring are generated by mutation only) or parents (if crossover is also used) and either ranks or selec
Data Loading...
