ON THE BOUNDARY OF DREAMS
With the recent advances in artificial intelligence and its application on art I asked myself the following questions:
Can art be automated? Can neural art be creative per se or is it bounded by human imagination?
The genius Dalí, would put a tin-plate on the floor and then sit by a chair beside it while holding a spoon with his hands over the plate. He would then try to fall asleep. Drifting between wakefulness and sleep, he would witness hallucinations and distorted versions of reality. At the moment of falling asleep, he would lose grip of the spoon and it would clang on the plate, waking him up and allowing him to capture his dream imagery.
In this work, Albert aims to explore the state of subconsciousness and the boundaries between sleep and wakefulness as a source of creativity. A custom generative adversarial network with attention mechanisms is used to generate patterns and shapes reminiscent of Dalí’s creative process. A set of psychedelic art was selected to induce the network into a state of surrealism and distortion. The output, an endless imaginative collection, is then blended to Dalinean style paintings using a multimodal style transfer network.
On the Boundary of Dreams was featured on the NeurIPS Workshop of Machine Learning for Creativity and Design in 2018.
Albert holds a B.Sc. and a M.Sc. in Electrical Engineering by the Polytechnic University of Catalonia with specialization in machine learning, computer vision and natural language processing. He was a founding member of the AI reading group and contributor in UPC deep learning seminars.
After a successful stay in the CSIRO research center in Australia, he moved to Toronto in 2018. There he worked at Triage (www.triage.com), designing and building software products able to recognize over 500 skin conditions with accuracy superior to dermatologists. He has been a teacher of the course “Machine Learning in Production” for the machine learning bootcamp AI Deep Dive and a distinguished speaker in dermatology conferences.
Albert is passionate about the intersection of machine learning with art, music and climate change and has authored multiple projects in topics such as music lyrics generation, image retrieval, music genre classification, and generative models for art paintings.