Research map / Probabilistic and scientific modeling / Molecules and proteins
Can I Have Your Order? Monte-Carlo Tree Search for Slot Filling Ordering in Diffusion Language Models
Joshua Ong Jun Leang, Yu Zhao, Mihaela C. Stoian, Wenda Li, Shay B Cohen, Eleonora Giunchiglia
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 MCTS slot ordering to ReFusion (ablation) | ReFusion 7B · HumanEval | Pass@1 (%): 62.05% -> 78.37% | Improved |
| add MCTS slot ordering to ReFusion (ablation) | ReFusion 7B · MBPP | Pass@1 (%): 54.12% -> 73.57% | Improved |
| add MCTS slot ordering to ReFusion (ablation) | ReFusion 7B · ARC | Pass@1 (%): 87.98% -> 88.68% | Improved |
| add MCTS slot ordering to ReFusion (ablation) | ReFusion 7B · MATH500 | Pass@1 (%): 42.90% -> 47.80% | Improved |
| replace sequential slot order with MCTS ordering (ablation) | McDiffuSE · Average of 6 benchmarks | Pass@1 (%): 58.29% -> 68.52% | Improved |
| replace random slot order with MCTS ordering (ablation) | McDiffuSE · Average of 6 benchmarks | Pass@1 (%): 43.10% -> 68.52% | Improved |
| increase simulations N_sim from 30 to 270 at low exploration c=2 (ablation) | McDiffuSE · MBPP and MATH500 | accuracy: mean entropy 1.41 -> 1.18 bits; accuracy decreases | Hurt |
| increase simulations N_sim from 30 to 270 at high exploration c=100 (ablation) | McDiffuSE · policy entropy | mean entropy (bits): 1.4176 -> 1.4340 | No clear effect |
| increase exploration constant c (ablation) | McDiffuSE · benchmarks | accuracy | Improved |
| raise N_sim from 30 to 270 (ablation) | McDiffuSE · accuracy and cost | accuracy / compute: improvement < 2% while computational cost increases nine-fold | Mixed |
| set tau=0.5 and lambda=0.3 versus lambda=0.0 (no confidence mixing) (ablation) | McDiffuSE · MATH500 | accuracy (%): 45.60 -> 48.20 | Improved |
| add MCTS to ReFusion versus Best-of-N and Beam Search at matched compute (ablation) | McDiffuSE · Average of 6 benchmarks (LM-Eval-Harness) | accuracy (%): Best-of-N 61.82, Beam Search 63.84 -> DiffuSearch 74.53 | Improved |
Closest papers by meaning
- The Flexibility Trap: Rethinking the Value of Arbitrary Order in Diffusion Language Models (ICML 2026 · similarity 0.88)
- Train for the Worst, Plan for the Best: Understanding Token Ordering in Masked Diffusions (ICML 2025 · similarity 0.88)
- Diffusion-State Policy Optimization for Masked Diffusion Language Models (NeurIPS 2026 · similarity 0.87)
- Monte Carlo Tree Diffusion for System 2 Planning (ICML 2025 · similarity 0.87)
- d2: Improving Reasoning in Diffusion Language Models via Trajectory Likelihood Estimation (ICML 2026 · similarity 0.86)
- UnMaskFork: Test-Time Scaling for Masked Diffusion via Deterministic Action Branching (ICML 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.
