Pub-AI: AI in Science

JEPA · 2026-09-30

D-JEPA: Design-Recoverable JEPA Representation with Swappable Physics Decoders

Nitin Nagesh Kulkarni, Aashwin Anand Mishra, Yin Yu, Peter Lyu

arXiv:2609.33110PDF

Auto-summarized: this summary was generated by a language model from the paper’s text and has not been reviewed by an editor. Check the paper before relying on it. Benchmark numbers appear only after a human has verified them.

TL;DR

D-JEPA learns a geometry-only latent representation and uses swappable, lightweight physics-specific decoders conditioned on operating conditions to predict physical fields. An explicit design-recoverability objective encourages the geometry latent to preserve underlying design variables for linear recovery and design optimization. The authors report that on four 3D benchmarks it maintains/improves full-field accuracy and achieves near-perfect linear recoverability while enabling decoder transfer.

Why it matters

Reusable geometry representations are needed for scientific surrogate modeling and design tasks where geometry changes, operating conditions vary, and physics response spaces differ. By factorizing a shared geometry latent from physics/condition-specific decoding, D-JEPA aims to support prediction accuracy, design-variable analysis/optimization, and reuse without retraining the geometry encoder.

Method

  • Geometry encoder Eθ maps a point cloud (surface + normals) to a compact geometry latent g (does not take operating conditions c or physics labels s).
  • Physics-specific FiLM-modulated implicit neural field decoders Ds(g,q,c) predict full fields; geometry latent is shared across conditions and regimes, decoders can be swapped with a frozen encoder.
  • Training includes JEPA latent prediction plus physical reconstruction, a design-recoverability loss via a linear probe, and case-level constraints to mitigate a collapse failure mode where targets become nearly invariant across geometries.

Limitation

The authors state the evaluation is limited to a small set of parameterized engineering datasets, cross-physics transfer is demonstrated in a single CFD-to-structural setting, operating-condition transfer is over a limited range of conditions, and practical engineering problems may involve constraints, discrete variables, noisy objectives, and stronger out-of-distribution conditions; broader multi-physics benchmarks and more diverse geometry families remain important for future evaluation.

Abstract (from arXiv)

Joint-Embedding Predictive Architectures (JEPAs) provide a framework for learning compact representations without directly reconstructing high-dimensional observations. However, in parameterized physical systems, learned representations can entangle geometry with operating conditions and task-specific physical responses, limiting their reuse across prediction tasks. We introduce D-JEPA (Design-recoverable JEPA), a geometry-centric JEPA that computes a compact representation from geometry alone and reuses it across operating conditions and physical response spaces through lightweight physics-specific decoders. An explicit design-recoverability objective encourages the geometry latent to preserve information about the underlying design variables, enabling the representation to support design analysis and optimization. We further identify a case-level collapse failure mode in which target representations become nearly invariant across distinct geometries despite low reconstruction error, and mitigate it using case-level variation constraints and auxiliary target reconstruction. Across four 3D aerodynamic, hydrodynamic, and structural benchmarks, D-JEPA maintains or improves full-field prediction accuracy while achieving near-perfect linear recoverability of design parameters. The frozen geometry representation can be reused at held-out operating conditions and transferred to a structural response task with fewer trainable parameters. Finally, the representation supports differentiable design optimization, with designs validated using high-fidelity CFD, preserving the predicted ranking of candidate designs. These results demonstrate that separating a reusable geometry representation from physics-specific prediction provides a practical representation for scientific surrogate modeling and design.

Related papers