Research map / Architectures and efficiency / RNNs and normalization
$\texttt{FlashSchNet}$: Fast and Accurate Coarse-Grained Neural Network Molecular Dynamics
Pingzhi Li, Hongxuan Li, Zirui Liu, Xingcheng Lin, Tianlong Chen
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 |
|---|---|---|---|
| replace CGSchNet with IO-aware fused kernels, CSR segment aggregation and 16-bit quantization (FlashSchNet) | CGSchNet · Homeodomain (1ENH), 64 replicas | speed (timestep*mol/s): 477 -> 3095 | Improved |
| use FlashSchNet's IO-aware design to avoid materializing edge tensors | CGSchNet · Homeodomain (1ENH), 64 replicas | peak memory (GB): 92.5 -> 18.0 | Improved |
| apply FlashSchNet to a small protein | CGSchNet · Chignolin | speed (timestep*mol/s): 3578 -> 5222 | Improved |
| use FlashSchNet on the largest protein | CGSchNet · Alpha3D | speed (timestep*mol/s): 288 -> 2610 | Improved |
| FlashSchNet optimizations versus CGSchNet on structural accuracy | FlashSchNet · Villin | GDT-TS: 0.78 -> 0.74 | No clear effect |
| FlashSchNet optimizations versus CGSchNet on native-contact fidelity | FlashSchNet · Villin | largest metastable Q: 0.96 -> 0.88 | Hurt |
| use learned potential (FlashSchNet) instead of MARTINI | MARTINI · Villin | GDT-TS: 0.46 -> 0.74 | Improved |
| grow the neighbor graph during unfolding (denser edges) for CGSchNet's scatter aggregation (ablation) | CGSchNet · 1ENH elongated 300k-step simulation | throughput | Hurt |
| use contention-free CSR segment reductions (ablation) | FlashSchNet · 1ENH elongated 300k-step simulation | throughput stability | Improved |
| reduce memory via IO-aware design to allow larger batch sizes (ablation) | FlashSchNet · replica scaling, four proteins | max replicas before OOM: 256 replicas (Alpha3D) to 2048 replicas (Chignolin) | Improved |
Closest papers by meaning
- FlashMD: long-stride, universal prediction of molecular dynamics (NeurIPS 2025 · similarity 0.88)
- FlashTP: Fused, Sparsity-Aware Tensor Product for Machine Learning Interatomic Potentials (ICML 2025 · similarity 0.88)
- Neural Network Reparametrization for Accelerated Optimization in Molecular Simulations (NeurIPS 2024 · similarity 0.87)
- Long-Short-Range Message-Passing: A Physics-Informed Framework to Capture Non-Local Interaction for Scalable Molecular Dynamics Simulation (ICLR 2024 · similarity 0.87)
- UniSim: A Unified Simulator for Time-Coarsened Dynamics of Biomolecules (ICML 2025 · similarity 0.86)
- ELoRA: Low-Rank Adaptation for Equivariant GNNs (ICML 2025 · 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.
