Neural Architecture Search‭. ‬Evolutionary Strategies in Art‭, ‬Design and Architecture‭. (‬evolution‭)‬

Arch‭. ‬Karolína Kotnour‭, ‬Czech Technical University in Prague‭ (‬CZ‭) ‬

Abstract‭: ‬This research proposes the Theory of Evolution for Architectural Intelligent Adaptive Systems‭. ‬The computational architectural language generates intelligent architectural structures that achieve balanced environmental conditions for individuals‭ ‬and communities based on their space experience‭, ‬sensing and behaviour‭. ‬The intelligence is encoded in a script of an algorithms‭ ‬of neural networks models that are capable of rewriting existing code protocols‭, ‬and therefore actively forward demands on architecture for effective and dynamic adaptability‭. ‬The major challenge faced by Architectural Intelligence is to develop agents that can learn to perform many different tasks and learn with the little domain-specific knowledge in real-time‭.‬

Hyper NEAT for Architectural Intelligence

Hyper NEAT is a generative incoming‭ ‬that continuously evolved ANN with principles of NEAT‭, ‬NeuroEvolution of Augmented Topologies‭. ‬The origin of life itself can be‭ ‬seen as a meta-evolutionary event‭. ‬The combination of ANN and EA is called NeuroEvolution‭ (‬NE‭), ‬where an evolution evolves a population of ANN‭. ‬This research proposes the Theory of Evolution for Architectural Intelligent Adaptive Systems.The Architectural‭ ‬Intelligence is a set of evolutionary‭ ‬mechanisms that has capability to adapt architectural organism to the dynamical environmental or behavioural situation‭. ‬The Architectural Intelligence is both adapting‭, ‬changing and accommodating the dynamic of the environment‭. ‬The Meta-Learning approach along with the NeuroEvolution algorithms of the large scale neural networks provide effective intelligent model for a continuous‭ ‬adaptation in a dynamic complex environments‭. ‬Neural Architecture Search‭ (‬NAS‭) ‬is the process of automating the network engineering‭. ‬The automatic prototyping and design of neural‭ ‬networks is based on NeuroEvolution that allies evolutionary algorithms to evolve network weights and topology‭.‬

Short CV‭: ‬artist/architect/director/researcher‭ ‬Karolína Kotnour‭ ‬is an architect and artist dedicated to an architectural spatial and audio-visual production‭. ‬She is focused on creating future‭ ‬evolving architecture by transforming‭ ‬methods from neuroscience‭, ‬machine learning‭, ‬immersive and sound spatialization research‭. ‬In her projects and installations‭, ‬she connects and synchronizes architectural and sound structures‭. ‬She claims‭ ‬‘the reciprocal confrontation of sound waves as is a liberated contour of space’‭. ‬She interests in‭ ‬‘space as evolving over time‭, ‬in parallels and mutual confrontations and reflections’‭. ‬A significant role plays human acoustic presence and performance‭. ‬She observes extreme space phenomena e.g‭.: ‬‮„‬acoustic black holes“‭ ‬and transformation of sound vibrations in their surroundings‭. ‬Karolína is an architect with international experience on projects of various scales and environmental contexts and cultural background‭. ‬She is Ph.D‭. ‬research fellow at FLOW studio at Faculty of Architecture CTU in Prague‭.‬