Research map / Probabilistic and scientific modeling / Molecules and proteins
Boosting World Models Learning via Latent-Space Value Alignment
Xingyu Jiang, Yuheng Pan, Mukang You, Xiuhui Zhang, Ning Gao, Guanwei Yan, Hao Li, Yue Deng
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 value-alignment regularization (Var) to DreamerV3 | DreamerV3 (2023 v1) · Atari 100k | human-normalized mean: 1.10 -> 1.34 | Improved |
| add value-alignment regularization (Var) to DreamerV3 | DreamerV3 (2023 v1) · Atari 100k | human-normalized median: 0.58 -> 1.00 | Improved |
| add value-alignment regularization (Var) to STORM | STORM · Atari 100k | human-normalized mean: 1.14 -> 1.36 | Improved |
| add value-alignment regularization (Var) to STORM | STORM · Atari 100k | human-normalized median: 0.51 -> 0.81 | Improved |
| add value-alignment regularization (Var) to DreamerV3 | DreamerV3 (2023 v1) · DMC visual (1M steps) | task mean score: 792 -> 827 | Improved |
| add value-alignment regularization (Var) to DreamerV3 | DreamerV3 (2023 v1) · DMC proprioceptive (500K steps) | task mean score: 805 -> 817 | Improved |
| add Var under 50% Gaussian noise in proprioceptive inputs | DreamerV3 (2023 v1) · Hopper Hop (50% noise) | score: 134 -> 187 | Improved |
| add Var under 50% Gaussian noise in proprioceptive inputs | DreamerV3 (2023 v1) · Acrobot Swingup (50% noise) | score: 124 -> 241 | Improved |
| add Var to STORM, measured on imagined rollouts | STORM · KungFuMaster | value error: 111.8 -> 38.9 | Improved |
| add Var to STORM, measured on imagined rollouts | STORM · KungFuMaster | reconstruction MSE: 3.2 × 10−4 -> 2.7 × 10−4 | Improved |
| add adaptive weighting (inverse dynamics loss) instead of static weight 0.5 (ablation) | STORM + Var · Atari 100k (Alien) | score: 987.9 -> 1361.4 | Improved |
| static weight 0.5 instead of no value alignment (ablation) | STORM · Atari 100k (Alien) | score: 1054.3 -> 987.9 | Hurt |
Closest papers by meaning
- Inverting the Bellman Equation: From $Q$-Values to World Models (NeurIPS 2026 · similarity 0.88)
- Simplifying Latent Dynamics with Softly State-Invariant World Models (NeurIPS 2024 · similarity 0.88)
- EDELINE: Enhancing Memory in Diffusion-based World Models via Linear-Time Sequence Modeling (NeurIPS 2025 · similarity 0.88)
- WALL-E: World Alignment by NeuroSymbolic Learning improves World Model-based LLM Agents (NeurIPS 2025 · similarity 0.88)
- HarmonyDream: Task Harmonization Inside World Models (ICML 2024 · similarity 0.88)
- Zero-shot World Models via Search in Memory (NeurIPS 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.
