Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-supervised Hierarchical Visual Reasoning with World Model
Yuanfei Xu, Lin Liu, Wengang Zhou, Mingxiao Feng, Houqiang Li
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 |
|---|---|---|---|
| extend ResDreamer from two to three hierarchy layers (ablation) | ResDreamer (50M) · MineDojo combat | success rate: 0.727 -> 0.776 | Improved |
| add the current layer's own rollout hint to residual hints (ablation) | ResDreamer (50Mx2) · MineDojo combat | success rate: 0.563 -> 0.727 | Improved |
| add residual connections to rollout hints (ablation) | ResDreamer (50Mx2) · MineDojo combat | success rate: 0.400 -> 0.727 | Improved |
| add hierarchy and residual connection to Dreamer with rollout foresight (ablation) | Dreamer (single layer) · MineDojo combat | success rate: 0.559 -> 0.727 | Improved |
| condition heads on stacked latent states of all layers instead of top layer only (ablation) | ResDreamer (50Mx2) · MineDojo combat | success rate: 0.727 -> 0.377 | Hurt |
| use longer foresight horizon and larger stride (H=4,8,16; D=1,2,4) (ablation) | ResDreamer · DMC Vision | average performance | Mixed |
| use ResDreamer 50Mx2 instead of DreamerV3 (fewer parameters) | DreamerV3 · MineDojo combat | average performance: 92.0M vs 109.5M parameters | Improved |
Closest papers by meaning
- UniVR: Thinking in Visual Space for Unified Visual Reasoning (NeurIPS 2026 · similarity 0.89)
- MindJourney: Test-Time Scaling with World Models for Spatial Reasoning (NeurIPS 2025 · similarity 0.87)
- DriveDreamer-Policy: A Geometry-Grounded World-Action Model for Unified Generation and Planning (NeurIPS 2026 · similarity 0.87)
- Thinking with Imagination: Agentic Visual Spatial Reasoning with World Simulators (NeurIPS 2026 · similarity 0.86)
- See Further, Think Deeper: Advancing VLM's Reasoning Ability with Low-level Visual Cues and Reflection (CVPR 2026 · similarity 0.86)
- MATA: A Trainable Hierarchical Automaton System for Multi-Agent Visual Reasoning (ICLR 2026 · 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.
