همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Development of 3D Neural Cellular Automata for Learning Spatio-Temporal Patterns

نویسندگان :

Yashar Rezazadeh Shahir¹ َُAkram Beigi²

1- دانشگاه تربیت دبیر شهید رجایی 2- دانشگاه تربیت دبیر شهید رجایی

کلمات کلیدی :

Neural Cellular Automata،3D Modeling،Evolutionary Computation،Machine Learning،Spatio-Temporal Pattern Processing

چکیده :

Neural Cellular Automata (NCA) have emerged as a powerful framework that integrates machine learning with mechanistic modeling to learn and reproduce complex spatio-temporal dynamics. Despite their promise, existing studies have largely focused on two-dimensional formulations, limiting their applicability to inherently volumetric phenomena. In this work, we introduce a three-dimensional extension of Neural cellular automata capable of learning, generating, and sustaining complex structures in 3D environments. Building upon the architecture of Richardson et al. (2024) in Learning Spatio-Temporal Patterns with Neural Cellular Automata, our model enables efficient training and robust generalization across a variety of volumetric patterns. Experimental results demonstrate that 3D NCA can synthesize intricate structures while maintaining computational efficiency. These findings broaden the potential of NCA-based systems and offer new opportunities in domains such as computational biology, medical image analysis, materials science, and artificial life.