Research Network Seminar 1st of February: generative AI and exploration of artistic styles
12 views
Skip to first unread message
Liubov
Jan 26, 2024, 12:19:57 PM1/26/24
to network-se...@googlegroups.com, rndapprendre
Dear all,
We are excited to share with you the research talk next week, which we continue as part of a (hyper:))-network seminar series this year. The first talk will be done by Levin
Brinkmann
(Max Planck).
Many interesting talks coming and some of them are online from last 2023 year (talk on embeddings, talk on Kreyon city and more)
Have a good year ahead!
Feel free to register if you are interested to get a link:
Title
Exploring the Evolution of Artistic Styles Using Generative AI
Abstract
Understanding the evolution of human creative expression is central to art history, and instrumental to the progress of algorithmic creativity. Recent advances in generative AI, such as Stable Diffusion, Midjourney, and DALL-E, show great promise in generating detailed images based on textual prompts blending visual concepts and art styles. However, whether these models can produce truly novel outputs beyond recombination remains unclear. Measuring creativity and cultural progress in subjective domains like the arts is challenging, but generative AI can help deconstruct art into distinct concepts, such as style, content, and composition, and measure their similarity. In this ongoing work, we propose a method to measure the similarity of visual concepts and thereby the cultural evolution of artist styles. We also present a simple influence model to represent the cultural processes, discuss the convex hull in embedding space as a measure of humanity's commutative cultural repertoire, and discuss the potential of generative AI to explore new, unseen art styles.
Exploring the Evolution of Artistic Styles Using Generative AI
Abstract
Understanding the evolution of human creative expression is central to art history, and instrumental to the progress of algorithmic creativity. Recent advances in generative AI, such as Stable Diffusion, Midjourney, and DALL-E, show great promise in generating detailed images based on textual prompts blending visual concepts and art styles. However, whether these models can produce truly novel outputs beyond recombination remains unclear. Measuring creativity and cultural progress in subjective domains like the arts is challenging, but generative AI can help deconstruct art into distinct concepts, such as style, content, and composition, and measure their similarity. In this ongoing work, we propose a method to measure the similarity of visual concepts and thereby the cultural evolution of artist styles. We also present a simple influence model to represent the cultural processes, discuss the convex hull in embedding space as a measure of humanity's commutative cultural repertoire, and discuss the potential of generative AI to explore new, unseen art styles.
Bio
Levin
Brinkmann, a Ph.D. candidate at the Max Planck Institute for Human
Development
under the supervision of Iyad Rahwan, blends complex systems, machine
learning, and experimental social science to study human-AI hybrid
systems. His non-academic work includes applying contrastive learning to
enhance the work of professional stylists. During
his Ph.D., he conducts experiments on behavior transmission between
deep Q-learning agents and humans, as well as on collaborative
exploration of
visual concepts using diffusion models. Levin's work is driven by his
passion for integrating insights from collective intelligence and
cultural evolution into AI, aiming to develop prosocial and innovative
AI ecosystems.
.jpg?part=0.1&view=1)
--
Reply all
Reply to author
Forward
0 new messages