Research map / Vision-language and multimodal / Multimodal LLMs
Latent Visual Diffusion Reasoning with Monte Carlo Tree Search
Xirui Teng, Nan Xi, Junsong Yuan
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 latent diffusion reasoning module (ablation) | LVDR · EgoExo4D | Spearman rho: 0.83 -> 0.88 | Improved |
| add latent diffusion reasoning module (ablation) | LVDR · EgoExo4D Soccer | Spearman rho: 0.97 -> 0.94 | Hurt |
| add latent diffusion reasoning module (ablation) | LVDR · FitnessAQA | F1: OK 0.768 -> 0.747; OE 0.497 -> 0.818 | Mixed |
| add latent diffusion reasoning module (ablation) | LVDR · JIGSAWS | Spearman rho: 0.65 -> 0.69 | Improved |
| add latent diffusion reasoning module (ablation) | LVDR · Cataract-101 | accuracy: 0.67 -> 0.71 | Improved |
| mask MCTS-selected keypoints (ablation) | LVDR · Cataract-101 | accuracy: 0.71 -> 0.53 | Hurt |
| mask MCTS-selected keypoints (ablation) | LVDR · EgoExo4D | Spearman rho: 0.88 -> 0.86 | No clear effect |
| mask MCTS-unselected keypoints (ablation) | LVDR · Cataract-101 | accuracy: 0.71 -> 0.71 | No clear effect |
| set loss weight lambda=1 instead of 0.5 (ablation) | LVDR · EgoExo4D | relative L2 (x100): 3.86 -> 1.69 | Improved |
| cosine noise schedule instead of linear (ablation) | LVDR · EgoExo4D | Spearman rho: 0.85 -> 0.88 | Improved |
| increase MCTS iterations from 75 to 225 (ablation) | LVDR · inference speed | latency per frame: 0.048s -> 0.129s | Hurt |
| use LVDR instead of MAGR++ | action quality assessment · JIGSAWS Needle-Passing | relative L2 (x100): 11.10 -> 18.28 | Hurt |
Closest papers by meaning
- SportR: A Benchmark for Multimodal Large Language Model Reasoning in Sports (ICLR 2026 · similarity 0.87)
- Leveraging Latent Visual Reasoning in Silence (NeurIPS 2026 · similarity 0.85)
- MARVEL: Multidimensional Abstraction and Reasoning through Visual Evaluation and Learning (NeurIPS 2024 · similarity 0.85)
- Hyperphantasia: A Benchmark for Evaluating the Mental Visualization Capabilities of Multimodal LLMs (NeurIPS 2025 · similarity 0.85)
- SLVR: Structured Latent Visual Reasoning via Human-like Reasoning Flows (NeurIPS 2026 · similarity 0.85)
- SoTA with Less: MCTS-Guided Sample Selection for Data-Efficient Visual Reasoning Self-Improvement (NeurIPS 2025 · 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.
