AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

PGD-NO: A Neural Operator with Precomputed Geometry Decomposition for 3D Million-Scale Physics Simulations

Weiheng Zhong, Jing Bi, Victor Oancea, Hadi Meidani

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Scalable approximation and equivariant compute · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add precomputed surface geometry tokens instead of grid-pooling tokens (ablation)PGD-NO-v1 · Heat SinkL2 relative error (%): 9.43 -> 2.47Improved
add precomputed surface geometry tokens instead of grid-pooling tokens (ablation)PGD-NO-v1 · JEBL2 relative error (%): 62.5 -> 37.2Improved
add precomputed surface geometry tokens instead of grid-pooling tokens (ablation)PGD-NO-v1 · AircraftL2 relative error (%): 35.4 -> 18.2Improved
add precomputed surface geometry tokens instead of grid-pooling tokens (ablation)PGD-NO-v1 · CFD-VOLL2 relative error (%): 36.7 -> 12.2Improved
increase geometry token layers from 2 to 12 (ablation)PGD-NO-v1 · CFD-VOLL2 relative error (%): 17.9 -> 11.1Improved
increase geometry token layers from 2 to 12 (ablation)Transolver++ · CFD-VOLL2 relative error (%): 17.8 -> 15.1Improved
increase number of geometry tokens from 32 to 64 (ablation)PGD-NO · JEBL2 relative error (%): 65.4 -> 41.5Improved
increase number of geometry tokens from 128 to 256 (ablation)PGD-NO · JEBL2 relative error (%): 37.2 -> 35.4Improved
add attention layer on geometry tokens (v2 vs v1)PGD-NO · JEBL2 relative error (%): 39.4 -> 37.2Improved
add attention layer on geometry tokens (v2 vs v1)PGD-NO · AircraftL2 relative error (%): 4.77 -> 5.37Hurt
use transposed segmentation matrix hierarchical mapping (v3 vs v1)PGD-NO · AircraftL2 relative error (%): 4.77 -> 4.66Improved
compare PGD-NO-v1 with Transolver++ at 12M points, full samplingPGD-NO-v1 · CFD-VOL (12M points)L2 relative error (%): 15.8 -> 12.2Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy