AtomWorld-Mem Restores Latent States for Long-Horizon Atomistic Simulation

AtomWorld-Mem uses spatial encoders and multi-scale keyframes to recover hidden dynamical contexts from instantaneous crystal snapshots. By integrating short-term event memory and long-term structural memory, the model transforms atomistic evolution into a memory-based world-state restoration problem for better future-predictive accuracy.