Research map / Probabilistic and scientific modeling / Molecules and proteins

Bridging Equivariant GNNs and Spherical CNNs for Structured Physical Domains

NeurIPS 2025 · Probabilistic and scientific modeling › Molecules and proteins · approach: Equivariant and disentangled latent representations

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Introduces G2Sphere, a method that maps object geometries to spherical signals by operating entirely in Fourier space to encode geometric structure into latent Fourier features.

Problem

Many modeling tasks (e.g., computing radar responses of objects, navigation in environments) can be framed as computing spherical signals from geometric inputs.

Approach

Maps object geometry to spherical signals entirely in Fourier space, encoding geometric structure into latent Fourier features; bridges equivariant GNNs and spherical CNNs.

Closest papers by meaning

• Scattering Networks on the Sphere for Scalable and Rotationally Equivariant Spherical CNNs (ICLR 2022 · similarity 0.89)
• Efficient Generalized Spherical CNNs (ICLR 2021 · similarity 0.88)
• PDO-eS2CNNs: Partial Differential Operator Based Equivariant Spherical CNNs (AAAI 2021 · similarity 0.88)
• Attention on the Sphere (NeurIPS 2025 · similarity 0.85)
• O$n$ Learning Deep O($n$)-Equivariant Hyperspheres (ICML 2024 · similarity 0.85)
• TetraSphere: A Neural Descriptor for O(3)-Invariant Point Cloud Analysis (CVPR 2024 · similarity 0.84)