Research map / Architectures and efficiency / CNN design
Scaling Laws and Symmetry, Evidence from Neural Force Fields
Khang Ngo, Siamak Ravanbakhsh
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 |
|---|---|---|---|
| use equivariant architecture (eSEN) instead of unconstrained MPNN | MPNN · force-field data scaling | data-scaling exponent beta: 0.311 -> 0.753 | Improved |
| use equivariant architecture (eSEN) instead of unconstrained MPNN | MPNN · force-field parameter scaling | parameter-scaling exponent alpha: 0.276 -> 0.817 | Improved |
| use higher-order equivariance (GemNet-OC) instead of EGNN | EGNN · force-field scaling | alpha: 0.387 -> 0.524 | Improved |
| use eSEN instead of unconstrained MPNN | MPNN · compute-optimal frontier | gamma_c: 0.142 -> 0.403 | Improved |
| add symmetry-regularization loss (ablation) | unconstrained MPNN · force-field data scaling | data exponent beta: slightly increases beta | Improved |
| add symmetry-regularization loss (ablation) | unconstrained MPNN · force-field parameter scaling | parameter exponent alpha: decreases alpha | Hurt |
| add symmetry-regularization loss (ablation) | unconstrained MPNN · compute-optimal slope | gamma_c: unchanged exponent; frontier shifted by 0.14 log10(6) | No clear effect |
| add data augmentation under 1% data for 100 epochs (ablation) | unconstrained MPNN · multi-epoch regime | compute-optimal power law: recovers gamma_c 0.14, F_c 0.96 | Improved |
| add test-time group averaging (M=32) (ablation) | unconstrained model · force-field parameter scaling | loss: minor change in constant A | No clear effect |
| scale equivariant vector channels in EGNN (ablation) | EGNN · validation loss | validation loss | Improved |
Closest papers by meaning
- A Tale of Two Symmetries: Exploring the Loss Landscape of Equivariant Models (NeurIPS 2025 · similarity 0.88)
- Parameter symmetries determine representational geometry in overparameterized nonlinear networks (NeurIPS 2026 · similarity 0.88)
- Analyzing Neural Scaling Laws in Two-Layer Networks with Power-Law Data Spectra (ICLR 2025 · similarity 0.88)
- Improving Equivariant Networks with Probabilistic Symmetry Breaking (ICLR 2025 · similarity 0.87)
- Symmetries in Overparametrized Neural Networks: A Mean Field View (NeurIPS 2024 · similarity 0.87)
- Neural Thermodynamics: Entropic Forces in Deep and Universal Representation Learning (NeurIPS 2025 · similarity 0.87)
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.
