The Flow has not summarised this story yet — read it at arXiv Machine Learning.
TRACE is a graph‑network simulator that stores granular contact history on edges using attention‑based message passing and a gated recurrent unit, with an edge‑identity dictionary to preserve memory as contacts change. It predicts normal and tangential forces while enforcing Coulomb friction and internal force balance, and is trained via single‑step pretraining followed by autoregressive fine‑tuning. On 2D and 3D granular column‑collapse benchmarks, TRACE achieves significantly lower long‑rollout position and deposit errors than existing simulators, uses fewer parameters, maintains near‑zero particle interpenetration, and outperforms the material point method by 12.2× (2D) and 8.9× (3D).
The study investigates granular slip forecasting by separating material state, loading progress, and window geometry effects using a compact neural score based on stress, pressure, coordination, non-affine motion, and force-network observables. Trained on 36 shear trajectories and tested on 18 new ones, the model ranked near‑slip windows better than prevalence or phase controls, yet loading‑history coordinates (e.g., causal elapsed strain) outperformed the compact score. The findings show that while compact observables contain temporally aligned slip information, stronger loading‑history baselines and geometry sensitivity limit the identification of a state‑specific short‑horizon precursor.
arXiv:2109.08813v1 Announce Type: cross Abstract: Seismic wave velocity of underground rock plays important role in detecting internal structure of the Earth. Rock physics models have long been the f...
arXiv:2607. 14233v1 Announce Type: cross Abstract: Physics-informed neural networks (PINNs) have had a broad research impact in modeling domains governed by partial differential equations (PDE).
arXiv:2608. 00593v1 Announce Type: cross Abstract: In stream finishing, the wear experienced by a workpiece depends strongly on its orientation within the rotating abrasive media.
Drift Field Net (DFN) is a deep neural network that predicts ocean surface flow fields from satellite observations, trained via a two‑stage strategy combining simulated data pretraining and Lagrangian fine‑tuning with an advection‑consistent loss. DFN improves particle trajectory forecasts, reducing mean positioning error by 20 km over a 7‑day period compared to an operational physics‑based model, and further decreasing error by 10 km when the advection loss is applied. The study demonstrates that incorporating Lagrangian constraints into deep‑learning training enhances ocean surface flow prediction accuracy.