Research map / Probabilistic and scientific modeling / Dynamical systems
Spectral-guided Physical Dynamics Distillation
Youjin Kim, Dagyeong Na, Jae Yong Lee, Junseok Kwon
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 SGDD distillation framework to EGNO decoder | EGNO · MD17 Benzene S2S | MSE x1e-2: 48.85 -> 12.97 | Improved |
| Add SGDD distillation framework to GFNODE decoder | GFNODE · MD17 Toluene S2S | MSE x1e-2: 4.82 -> 5.16 | Hurt |
| Add SGDD to EGNO decoder on motion capture | EGNO · Motion capture Walk S2S | MSE x1e-2: 8.1 -> 6.7 | Improved |
| Add SGDD to EGNO on ADk protein | EGNO · ADk protein S2S | MSE: 2.23 -> 1.75 | Improved |
| Add spectral-guided enhancement (SGE) (ablation) | SGDD-EGNO · MD17 Toluene S2T | MSE x1e-2: 4.18 -> 3.80 | Improved |
| Add feature alignment (ablation) | SGDD-EGNO · MD17 Toluene S2T | MSE x1e-2: 4.86 -> 3.80 | Improved |
| Add frequency alignment (ablation) | SGDD-EGNO · MD17 Toluene S2T | MSE x1e-2: 4.65 -> 3.80 | Improved |
| Add SGE (ablation) | SGDD-EGNO · Mocap-Walk S2T | MSE x1e-2: 4.04 -> 2.95 | Improved |
| Add SGE (ablation) | SGDD-EGNO · Mocap-Run S2T | MSE x1e-2: 12.61 -> 12.98 | Hurt |
| Add learnable weights (SGE) to SGDD-GFNODE (ablation) | SGDD-GFNODE · MD17 Benzene S2S | MSE x1e-2: 3.50 -> 2.23 | Improved |
| Truncation K=64 instead of K=16 (ablation) | SGDD-EGNO · Motion capture Walk S2S | MSE x1e-2: 9.79 -> 6.74 | Improved |
| Truncation K=64 instead of full basis (155) (ablation) | SGDD-EGNO · Motion capture Walk S2S | MSE x1e-2: 7.26 -> 6.74 | Improved |
Closest papers by meaning
- FLD: Fourier Latent Dynamics for Structured Motion Representation and Learning (ICLR 2024 · similarity 0.86)
- ODEWorld: A Continuous Predictive Architecture via Physical-Time Flow (NeurIPS 2026 · similarity 0.85)
- Predicting the Energy Landscape of Stochastic Dynamical System via Physics-informed Self-supervised Learning (ICLR 2025 · similarity 0.85)
- Generating Physical Dynamics under Priors (ICLR 2025 · similarity 0.85)
- Equivariant Graph Neural Operator for Modeling 3D Dynamics (ICML 2024 · similarity 0.84)
- Geometric Trajectory Diffusion Models (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.
