Research map / Reinforcement learning / RL from feedback
Discrete Tilt Matching
Yuyuan Chen, Shiyi Wang, Peter Potaptchik, Jaeyeon Kim, Michael Samuel Albergo
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 |
|---|---|---|---|
| use control variate c=1 with smaller annealing step (ablation) | DTM · Maze planning | reward/validity/diversity | Improved |
| remove control variate (c=0) (ablation) | DTM · Maze planning | mode collapse | Hurt |
| use overly large annealing step h (ablation) | DTM · Maze planning | mode collapse | Hurt |
| set annealing step h=6 vs {1.5,3,12} (ablation) | DTM · Countdown | accuracy | Improved |
| moderate h preserves multiplication/division mode (ablation) | DTM vs SPG · Countdown multiplication/division mode | accuracy: 1.95% (base) -> 2.78% (DTM); SPG 0% | Improved |
| fine-tune with DTM | LLaDA-8B-Instruct · Sudoku (3-shot) | accuracy: 27.7 -> 99.2 (len 256) | Improved |
| fine-tune with DTM | LLaDA-8B-Instruct · Countdown (0-shot) | accuracy: 16.8 -> 81.3 (len 256) | Improved |
| fine-tune with DTM | LLaDA-8B-Instruct · MATH500 (0-shot) | accuracy: 32.4 -> 36.0 (len 256); SPG 40.0 | Mixed |
| fine-tune with DTM | LLaDA-8B-Instruct · GSM8K (0-shot) | accuracy: 77.2 -> 81.6 (len 256); SPG 86.1 | Mixed |
| increase generation length 256 to 512 (ablation) | DTM · MATH500 / GSM8K | accuracy: 36.0 -> 40.2; 81.6 -> 83.2 | Improved |
| use DTM instead of SPG | Sudoku training · Sudoku reward vs wallclock | reward | Improved |
Closest papers by meaning
- d1: Scaling Reasoning in Diffusion Large Language Models via Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Revolutionizing Reinforcement Learning Framework for Diffusion Large Language Models (ICLR 2026 · similarity 0.88)
- d2: Improving Reasoning in Diffusion Language Models via Trajectory Likelihood Estimation (ICML 2026 · similarity 0.87)
- Reinforcement Learning for Diffusion LLMs with Entropy-Guided Step Selection and Stepwise Advantages (NeurIPS 2026 · similarity 0.87)
- Break the Block: Dynamic-size Reasoning Blocks for Diffusion Large Language Models via Monotonic Entropy Descent with Reinforcement Learning (ICML 2026 · similarity 0.87)
- Trajectory Balance with Asynchrony: Decoupling Exploration and Learning for Fast, Scalable LLM Post-Training (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.
