ABSTRACT High fidelity dynamic FE simulation is indispensable in product development, yet its computational cost becomes prohibitive when exploring diverse design candidates. Model order reduction (MOR) mitigates this cost by projecting the governing equations onto a low dimensional subspace, and nonlinear problems typically require additional hyperreduction to reduce the cost of evaluating internal forces. However, conventional MOR pipelines are strongly tied to a fixed number of degrees of freedom and mesh topology, which limits their applicability to non-parametric geometry variations. This work investigates a Neural Network–ROM strategy [1] for nonlinear hyperelastic dynamics in which a Neural Operator predicts a geometry-dependent solution subspace, and a reduced order model is then solved in that subspace. To avoid basis vectors (mode) misalignment under mode switching between different geometries, we employ learning at the subspace level using Grassmannian losses [1-3]. Moreover, we analyze the coupling of the predicted subspace with the geometry-conditioned hyperreduction DEIM [4] operators, with the aim of achieving efficient online evaluation in nonlinear dynamics. The presentation will report on the current implementation and discuss accuracy-cost trade-offs and generalization across geometry families. REFERENCES [1] Matray, Victor, David Néron, Frédéric Feyel, and Faisal Amlani. 2026. “Geometry-Agnostic Model Reduction with GNN-Generated Reduced POD Bases and Boosted PGD Enrichment for (Non)Linear Structural Elastodynamics.” Computer Methods in Applied Mechanics and Engineering 448 (118357): 118357. [2] Fanaskov, Vladimir, Vladislav Trifonov, Alexander Rudikov, Ekaterina Muravleva, and Ivan Oseledets. 2025. “Deep Learning for Subspace Regression.” arXiv [Cs.LG]. arXiv. http://arxiv.org/abs/2509.23249. [3] Zhang, Jiayao, Guangxu Zhu, Robert W. Heath Jr, and Kaibin Huang. 2018. “Grassmannian Learning: Embedding Geometry Awareness in Shallow and Deep Learning.” arXiv [Cs.LG]. arXiv. http://arxiv.org/abs/1808.02229. [4] Chaturantabut, S., & Sorensen, D. C. (2010). Nonlinear model reduction via discrete empirical interpolation. SIAM Journal on Scientific Computing, 32(5), 2737-2764.