Original title: End-to-end Phase Field Model Discovery Combining Experimentation, Crowdsourcing, Simulation and Learning
Authors: Md Nasim, Anter El-Azab, Xinghang Zhang, Yexiang Xue
The article addresses challenges in AI-driven scientific discovery due to large-scale experiment data. It introduces Phase-Field-Lab, a platform merging experimentation, crowdsourcing, simulation, and learning to discover phase field physics models automatically. It pioneers a streamlined annotation tool, cutting annotation time by ~50-75% while enhancing accuracy. Utilizing an end-to-end neural model, it embeds phase field simulation and domain knowledge, enabling automatic learning from data. The platform offers novel interfaces and visualizations fostering collaboration among experimental physicists, theorists, and computer scientists. Deployed in nano-structure evolution analysis under extreme conditions, it unveils new properties of nano-void defects, unattainable through manual methods. Phase-Field-Lab represents a breakthrough, revolutionizing scientific discovery by automating model discovery from vast experiment data and fostering interdisciplinary collaboration.
Original article: https://arxiv.org/abs/2311.12801