Research map / Vision-language and multimodal / Multimodal LLMs
Constructing and Interpreting Digital Twin Representations for Visual Reasoning via Large Language Models and Reinforcement Learning
Yiqing Shen, Mathias Unberath
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 format rewards (ablation) | DT-R1 · JiTBench Level 2 | J: 0.412 -> 0.808 | Improved |
| add DAG validation reward (ablation) | DT-R1 · JiTBench Level 2 | J: 0.687 -> 0.808 | Improved |
| add code execution reward (ablation) | DT-R1 · JiTBench Level 2 | J: 0.724 -> 0.808 | Improved |
| add accuracy rewards (ablation) | DT-R1 · JiTBench Level 2 | J: 0.523 -> 0.808 | Improved |
| add format rewards on top of task accuracy only (ablation) | DT-R1 · JiTBench Level 2 | J: 0.298 -> 0.808 | Improved |
| add format rewards on training success rate (ablation) | DT-R1 · JiTBench Level 2 | training success rate: 31.5% -> 94.2% | Improved |
| add SAM2 specialist (ablation) | DT-R1 · JiTBench Level 2 | J: 0.684 -> 0.808 | Improved |
| add OWLv2 specialist (ablation) | DT-R1 · JiTBench Level 2 | J: 0.732 -> 0.808 | Improved |
| add DepthAnything specialist (ablation) | DT-R1 · JiTBench Level 2 | J: 0.756 -> 0.808 | Improved |
| add Qwen2.5-VL specialist (ablation) | DT-R1 · JiTBench Level 2 | J: 0.778 -> 0.808 | Improved |
| use Qwen2.5-7B backbone instead of DeepSeek-R1-Distill-Qwen-7B (ablation) | DT-R1 · JiTBench Level 2 | J: 0.808 -> 0.798 | No clear effect |
| use PPO instead of GRPO | DT-R1 · JiTBench Level 2 | J: 0.808 -> 0.794 | No clear effect |
Closest papers by meaning
- OneThinker: All-in-one Reasoning Model for Image and Video (CVPR 2026 · similarity 0.91)
- Latent Visual Reasoning (ICLR 2026 · similarity 0.91)
- Multi-step Visual Reasoning with Visual Tokens Scaling and Verification (NeurIPS 2025 · similarity 0.90)
- More Thought, Less Accuracy? On the Dual Nature of Reasoning in Vision-Language Models (ICLR 2026 · similarity 0.90)
- MiCo: Multi-image Contrast for Reinforcement Visual Reasoning (NeurIPS 2025 · similarity 0.90)
- Reason-RFT: Reinforcement Fine-Tuning for Visual Reasoning of Vision Language Models (NeurIPS 2025 · similarity 0.90)
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.
