AI Paper Atlas

Research map / Probabilistic and scientific modeling / Differential equations and PDEs

From Cheap Geometry to Expensive Physics: Elevating Neural Operators via Latent Shape Pretraining

Zhizhou Zhang, Youjia Wu, Kaixuan Zhang, Yanjia Wang

ICLR 2026 · Probabilistic and scientific modeling › Differential equations and PDEs · approach: Neural operators for general geometries

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
feed stage-1 pretrained occupancy VAE latents to the operator instead of raw point cloudsGNOT · Electrostatics, mesh queryrelative L2 (x10^-2): 4.2 -> 3.3Improved
feed stage-1 pretrained latents instead of raw point cloudsLNO · Stress, mesh queryrelative L2 (x10^-2): 26.5 -> 13.6Improved
feed stage-1 pretrained latents instead of raw point cloudsTransolver · AirfRans (near), random queryrelative L2 (x10^-2): 15.0 -> 10.8Improved
feed stage-1 pretrained latents instead of raw point cloudsGNOT · Inductor (3D), mesh queryrelative L2 (x10^-2): 7.0 -> 7.1No clear effect
feed stage-1 pretrained latents instead of raw point cloudsLNO · AirfRans (near), mesh queryrelative L2 (x10^-2): 27.4 -> 27.1No clear effect
use random queries instead of mesh queriestwo-stage framework · all four datasetssize of improvementImproved
pretrain VAE on more geometry data (VAE3 vs VAE1) (ablation)GNOT with VAE latents · Stress, random queryrelative L2 (x10^-2): 9.1 -> 8.3Improved
pretrain VAE only on the same geometries as stage 2 (VAE1) (ablation)Transolver · Stress, mesh queryrelative L2 (x10^-2): 11.5 -> 12.0Hurt
lower KL weight from 0.001 to 0.0001 (ablation)GNOT with VAE latents · AirfRans (near), mesh queryrelative L2 (x10^-2): 5.6 -> 5.3Improved
use deterministic autoencoder instead of VAE (ablation)Transolver · AirfRans (near), mesh queryrelative L2 (x10^-2): 12.7 -> 11.3Improved
use shortest-vector proxy task instead of occupancy (ablation)GNOT · AirfRans, mesh queryrelative L2 (x10^-2): 5.6 -> 4.8Improved
use SDF proxy task pretraining with mesh query (ablation)GNOT · AirfRans, mesh queryrelative L2 (x10^-2): 5.2 -> 5.3No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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