EvoSim: Learning to Model, Modeling to Learn
An AI scientist that builds models of the microscopic world. It describes states with physical models, lets experimental discrepancies drive revisions to mechanisms and equations, and improves itself recursively. It reaches state-of-the-art accuracy on two industrial battery modeling and prediction tasks, surpassing models built by human experts.
1.79%Lithium-plating onset error (MAE, SOC)
7.62 mVVoltage error under vehicle driving (RMSE)
−36%Error frequency after self-evolution
