#machine learning
May 2026
AeroJEPA: Learning Semantic Latent Representations for Scalable 3D Aerodynamic Field Modeling
In this selective-decoding experiment, AeroJEPA substantially outperforms the evaluated direct-regression surrogates while avoiding predictions over the remainder of the aircraft, and shows how predictive representations can support aerodynamic analysis with or without full-field reconstruction.
Francisco Giral, A. Vishwasrao, Andrea Arroyo Ramo et al.
· arXiv.org · 1 citation