Research map / Probabilistic and scientific modeling / Molecules and proteins
Coordinate Denoising for Non-Equilibrium Molecular Representation Learning
Qianwei Tang, Baile Xu, Jian Zhao, Furao Shen
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 |
|---|---|---|---|
| add NDeM auxiliary denoising task (ablation) | TorchMD-NET · MD17 Aspirin force | force MAE: 0.2087 -> 0.1654 | Improved |
| raise auxiliary weight from 1 to 100 (ablation) | NDeM · MD17 Aspirin force | force MAE: 0.1654 -> 0.1722 | Hurt |
| increase max noise sigma from 0.05 to 0.5 (ablation) | NDeM · MD17 Aspirin force | force MAE: 0.1654 -> 0.1887 | Hurt |
| add NDeM when fine-tuning from Frad pretrained weights (ablation) | Frad · MD17 Aspirin force | force MAE: 0.2088 -> 0.1774 | Improved |
| add NDeM when fine-tuning from SliDe pretrained weights (ablation) | SliDe · MD17 Aspirin force | force MAE: 0.1698 -> 0.1648 | Improved |
| output noisy force through the primary force head instead of a separate auxiliary head (ablation) | NDeM · MD17 | force MAE | Hurt |
| add NDeM to EquiformerV2 (lambda_E=2) | EquiformerV2 · OC20 S2EF test | force MAE (meV/A): 13.8 -> 12.9 | Improved |
| add NDeM to EquiformerV2 (lambda_E=4) | EquiformerV2 · OC20 S2EF test | force MAE (meV/A): 14.2 -> 13.8 | Improved |
| NDeM instead of DeNS on EquiformerV2 (lambda_E=4) | EquiformerV2 · OC20 S2EF test | force MAE (meV/A): 13.4 (DeNS) vs 13.8 (NDeM) | No clear effect |
| add NDeM as auxiliary task to TorchMD-NET without pre-training | TorchMD-NET · QM9 gap | MAE (meV): 26.2 (SliDe) -> 26.0 | Improved |
| NDeM vs SliDe on QM9 U0 | NDeM · QM9 U0 | MAE (meV): 4.28 (SliDe) vs 5.31 (NDeM) | Hurt |
| NDeM on MD17 Benzene and Uracil vs best baselines | TorchMD-NET · MD17 | force MAE: Benzene 0.1802 vs 0.1691 (SliDe); Uracil 0.0806 vs 0.0760 (Frad) | Mixed |
Closest papers by meaning
- Learning 3D Anisotropic Noise Distributions Improves Molecular Force Fields (NeurIPS 2025 · similarity 0.90)
- Geometric-Facilitated Denoising Diffusion Model for 3D Molecule Generation (AAAI 2024 · similarity 0.88)
- Sliced Denoising: A Physics-Informed Molecular Pre-Training Method (ICLR 2024 · similarity 0.88)
- 3D Denoisers Are Good 2D Teachers: Molecular Pretraining via Denoising and Cross-Modal Distillation (AAAI 2025 · similarity 0.87)
- SubgDiff: A Subgraph Diffusion Model to Improve Molecular Representation Learning (NeurIPS 2024 · similarity 0.87)
- The dark side of the forces: assessing non-conservative force models for atomistic machine learning (ICML 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.
