Research map / Graph learning / GNN architectures
Spherical Channels for Modeling Atomic Interactions
C. Lawrence Zitnick, Abhishek Das, Adeesh Kolluru, Janice Lan, Muhammed Shuaibi, Anuroop Sriram, Zachary Ward Ulissi, Brandon M Wood
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 |
|---|---|---|---|
| Remove rotation of messages before and after the spherical-channel transform (ablation) | SCN · OC20 2M validation | Force MAE: 67.7 vs 24.6 | Improved |
| Replace non-linear aggregation with simple summation (No 1x1 conv) (ablation) | SCN m in [-1,1] · OC20 2M validation | Force MAE: 26.2 vs 24.6 | Improved |
| Use m in [-1,1] instead of only m=0 in message passing (ablation) | SCN · OC20 2M validation | Force MAE: 26.5 -> 24.6 | Improved |
| Increase orders to m in [-2,2] (ablation) | SCN · OC20 2M validation | Force MAE: 24.6 -> 23.4 | Improved |
| Estimate forces as energy gradients (grad-forces) instead of direct output (ablation) | SCN L=4 · OC20 2M validation | Force MAE: 26.2 vs 25.3 | Hurt |
| Increase L from 2 to 8 (ablation) | SCN · OC20 2M validation | Force MAE: 27.6 -> 23.2 | Improved |
| Use 4-tap sampling of z-axis rotations (ablation) | SCN L=6 16 layers · OC20 2M validation | Force MAE: 24.3 -> 22.7 | Improved |
| Use two bands (L and L-1) when aggregating messages (ablation) | SCN · OC20 2M validation | Energy MAE: 299 -> 292 | Improved |
| Combine 4-tap and 2-band (ablation) | SCN · OC20 2M validation | Energy MAE: 299 -> 279 | Improved |
| Use 4-taps (ablation) | SCN · OC20 2M validation | Samples / GPU sec: 7.7 -> 3.7 | Hurt |
| Increase depth from 12 to 16 layers (ablation) | SCN m=0 L=6 · OC20 2M validation | Energy MAE: 307 -> 300 | Improved |
| Scale training data from 2M to All+MD with larger SCN (ablation) | SCN L=8 K=20 · OC20 test IS2RE | Energy MAE: 330 -> 321 | Improved |
Closest papers by meaning
- The dark side of the forces: assessing non-conservative force models for atomistic machine learning (ICML 2025 · similarity 0.86)
- Learning the Electronic Hamiltonian of Large Atomic Structures (ICML 2025 · similarity 0.85)
- Neural Atoms: Propagating Long-range Interaction in Molecular Graphs through Efficient Communication Channel (ICLR 2024 · similarity 0.85)
- Pushing the Limits of All-Atom Geometric Graph Neural Networks: Pre-Training, Scaling, and Zero-Shot Transfer (ICLR 2025 · similarity 0.85)
- Higher-Rank Irreducible Cartesian Tensors for Equivariant Message Passing (NeurIPS 2024 · similarity 0.85)
- The Importance of Being Scalable: Improving the Speed and Accuracy of Neural Network Interatomic Potentials Across Chemical Domains (NeurIPS 2024 · similarity 0.84)
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.
