Research map / Label-efficient and robust learning / Self-supervised and contrastive
DreamSAC: Learning Hamiltonian World Models via Symmetry Exploration
Jinzhou Tang, Fan Feng, Minghao Fu, Wenjun Lin, Jing Yang, Biwei Huang, Keze Wang
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 viewpoint-robustness contrastive loss (ablation) | DreamSAC · Reacher (Unseen View) | mean reward: 212.37 -> 321.90 | Improved |
| use G-invariant Lie Transformer Hamiltonian instead of MLP (ablation) | DreamSAC · Walker (1.5x Gravity) | MSE: 4.9673 -> 1.0044 | Improved |
| add SAVi object-centric encoder (ablation) | DreamSAC · Avg. OOD | MSE: 0.9031 -> 0.7048 | Improved |
| replace random exploration with Symmetry Exploration (ablation) | DreamSAC · FetchPush (H=8) | image prediction MSE: 0.675 -> 0.302 | Improved |
| replace random exploration with Symmetry Exploration (ablation) | DreamSAC · Cheetah (H=16) | image prediction MSE: 0.1565 -> 0.4052 | Hurt |
| use Symmetry Exploration instead of RND exploration (with DreamerV3 world model swapped for Hamiltonian model) | DreamSAC · FetchPush (H=8) | image prediction MSE: 0.976 -> 0.302 | Improved |
| replace DreamerV3 with DreamSAC | DreamerV3+RND · Walker-walk (Unseen Gravity 1.5x) | mean reward: 167.52 -> 499.91 | Improved |
| replace DreamerV3 with DreamSAC | DreamerV3+Policy · Cheetah-run (Unseen Friction 2.0x) | mean reward: 118.79 -> 120.23 | No clear effect |
| replace DreamerV3+RND with DreamSAC | DreamerV3+RND · Hopper-hop (ID) | mean reward: 389.142 -> 366.117 | Hurt |
| replace DreamerV3 with DreamSAC | DreamerV3+Policy · Walker-walk (ID) | mean reward: 965.217 -> 996.502 | Improved |
| zero-shot (frozen) DreamSAC instead of fine-tuned (ablation) | DreamSAC · Reacher-hard (Unseen View) | mean reward: 321.90 -> 149.64 | Hurt |
Closest papers by meaning
- Hamiltonian Score Matching and Generative Flows (NeurIPS 2024 · similarity 0.86)
- Towards Cross Domain Generalization of Hamiltonian Representation via Meta Learning (ICLR 2024 · similarity 0.85)
- Latent Mixture of Symmetries for Sample-Efficient Dynamic Learning (NeurIPS 2025 · similarity 0.85)
- SymDrift: One-Shot Generative Modeling under Symmetries (NeurIPS 2026 · similarity 0.85)
- Physics-Informed Regularization for Domain-Agnostic Dynamical System Modeling (NeurIPS 2024 · similarity 0.85)
- Latent Intuitive Physics: Learning to Transfer Hidden Physics from A 3D Video (ICLR 2024 · similarity 0.85)
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.
