Research map / Probabilistic and scientific modeling / Dynamical systems
From Embedding to Control: Representations for Stochastic Multi-Object Systems
Xiaoyuan Cheng, Yiming Yang, Wei Jiang, Chenyang Yuan, Zhuo Sun, Yukun Hu
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 uniform neighbor weights (Hom) with mean-field adaptive weights (Hom+Mean) (ablation) | GCE · Rope few-shot, 1 trajectory | control error: 0.74 -> 0.51 | Improved |
| replace Dense pair-wise operators with Hom+Mean (ablation) | GCE · Rope few-shot, 1 trajectory | control error: 0.79 -> 0.51 | Improved |
| Hom+Mean versus Dense with 32 training trajectories (ablation) | GCE · Rope few-shot, 32 trajectories | control error: 0.26 -> 0.23 | Improved |
| use Gaussian potential instead of Laplace for interaction weights (ablation) | GCE · Swim few-shot | control cost: 435.7 -> 404.3 | Improved |
| use Gaussian potential instead of vMF for interaction weights (ablation) | GCE · Swim few-shot | control cost: 452.3 -> 404.3 | Improved |
| use a neural network (MLP) for the energy function instead of Gaussian kernel (ablation) | GCE · multiple environments | stability | Hurt |
| set Gaussian kernel bandwidth sigma to 2 rather than other values in 0.1 to 10 (ablation) | GCE (Gaussian) · Rope and Soft few-shot | control error and cost | Improved |
| set feature dimension to 32 (ablation) | GCE · Rope and Soft | control error | Improved |
| use characteristic kernel features instead of polynomial features (KPM) | GCE vs KPM · Power-Grid with 5% noise | control error: KPM NaN -> 0.39 ± 0.024 | Improved |
| use non-uniform mean-field weights instead of uniform weights (CKO) | GCE vs CKO · Swim few-shot | control cost: 421.0 -> 404.3 | Improved |
| use GCE instead of CKO on in-distribution Swim | GCE vs CKO · Swim in-distribution | control cost: 389.1 -> 383.7 | No clear effect |
| increase observation noise to 20% in Power-Grid | GCE vs CKO · Power-Grid 20% noise | control error: 0.85 -> 0.83 | No clear effect |
Closest papers by meaning
- Graph Mamba Operator: A Latent Simulator for Interacting Particle Systems (NeurIPS 2026 · similarity 0.86)
- PGODE: Towards High-quality System Dynamics Modeling (ICML 2024 · similarity 0.85)
- Physics-Informed Representation and Learning: Control and Risk Quantification (AAAI 2024 · similarity 0.85)
- Characterizing control between interacting subsystems with deep Jacobian estimation (NeurIPS 2025 · similarity 0.84)
- SEGNO: Generalizing Equivariant Graph Neural Networks with Physical Inductive Biases (ICLR 2024 · similarity 0.84)
- A Graph Dynamics Prior for Relational Inference (AAAI 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.
