Research map / Point clouds / Point cloud classification
Smooth, exact rotational symmetrization for deep learning on point clouds
Sergey Pozdnyakov, Michele Ceriotti
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 changed | Where | Measured | Effect |
|---|---|---|---|
| shrink PET from d_PET=256 to 128 (ablation) | PET · COLL | MAE force (meV/A): 23.1 -> 29.8 | Hurt |
| shrink PET from d_PET=256 to 128 (ablation) | PET · COLL | MAE energy (meV/molecule): 12.0 -> 15.3 | Hurt |
| use rotationally invariant 2-body PET instead of PET-128 (ablation) | PET-128 · COLL | MAE force (meV/A): 29.8 -> 56.7 | Hurt |
| apply ECSE symmetrization to PET-256 | PET-256 · COLL | MAE energy (meV/molecule): 12.0 -> 11.9 | No clear effect |
| apply ECSE symmetrization to PET | PET · HME21 | MAE energy (meV/atom): 17.8 -> 17.8 | No clear effect |
| average 5 PET models (ensemble) with ECSE | PET-ECSE · HME21 | MAE |f| (meV/A): 141.6 -> 128.5 | Improved |
| average predictions over 10 random orientations (PET-10) instead of one | PET-1 · HME21 | MAE energy (meV/atom): 17.8 -> 17.7 | No clear effect |
| use real-valued spins as atomic labels instead of discretized (ablation) | PET-ECSE · MnO | MAE energy (meV): 0.30 -> 0.14 | Improved |
| remove spin information (ablation) | PET-ECSE · MnO | MAE energy (meV): 0.30 -> 11.6 | Hurt |
| use cutoff 4.25 A instead of 3.7 A (ablation) | PET · water | MAE force (meV/A): 14.4 | Improved |
| 6x2 (blocks x transformer layers) instead of 12x1 stack (ablation) | PET · water | force MAE | Improved |
| add more GNN blocks / deeper transformers (ablation) | PET · water | force MAE | Improved |
Closest papers by meaning
- Scalable Non-Equivariant 3D Molecule Generation via Rotational Alignment (ICML 2025 · similarity 0.86)
- Symphony: Symmetry-Equivariant Point-Centered Spherical Harmonics for 3D Molecule Generation (ICLR 2024 · similarity 0.86)
- SymDrift: One-Shot Generative Modeling under Symmetries (NeurIPS 2026 · similarity 0.86)
- EquiformerV2: Improved Equivariant Transformer for Scaling to Higher-Degree Representations (ICLR 2024 · similarity 0.86)
- SE(3)-bi-equivariant Transformers for Point Cloud Assembly (NeurIPS 2024 · similarity 0.86)
- GeoMFormer: A General Architecture for Geometric Molecular Representation Learning (ICML 2024 · similarity 0.86)
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.
