Research map / Reinforcement learning / RL from feedback
Discrete Diffusion Trajectory Alignment via Stepwise Decomposition
Jiaqi Han, Austin Wang, Minkai Xu, Wenda Chu, Meihua Dang, Haotian Ye, Huayu Chen, Yisong Yue, Stefano Ermon
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 |
|---|---|---|---|
| increase samples per prompt N from 2 (pairwise) to 25 (ablation) | SDPO · Protein inverse folding | Pred-ddG: 0.529 -> 1.119 | Improved |
| increase N from 25 to 100 (ablation) | SDPO · Protein inverse folding | Pred-ddG: 1.119 -> 1.061 | No clear effect |
| lower beta (weaker KL regularization) (ablation) | SDPO · DNA design / protein folding | Pred-Activity / Pred-ddG | Mixed |
| add iterative labeling (2 rounds) (ablation) | SDPO · DNA design | Pred-Activity | Improved |
| SDPO vs DRAKES | DNA design · DNA enhancer design (HepG2) | Pred-Activity: 5.61 -> 6.30 | Improved |
| SDPO vs diffu-GRPO | protein inverse folding · Protein inverse folding | Success rate (%): 37.2 -> 75.5 | Improved |
| SDPO vs DRAKES | protein inverse folding · Protein inverse folding | Success rate (%): 78.6 -> 75.5 | Hurt |
| apply SDPO to LLaDA-8B-Instruct | LLaDA-8B-Instruct · GSM8K | accuracy: 78.6 -> 81.2 | Improved |
| D2-DPO on LLaDA-8B-Instruct | LLaDA-8B-Instruct · GSM8K | accuracy: 78.6 -> 78.1 | No clear effect |
| SDPO vs on-policy training cost | DRAKES · DNA design | seconds per step: 6.02 -> 0.77 | Improved |
Closest papers by meaning
- Fine-Tuning Discrete Diffusion Models via Reward Optimization with Applications to DNA and Protein Design (ICLR 2025 · similarity 0.90)
- Steering Masked Discrete Diffusion Models via Discrete Denoising Posterior Prediction (ICLR 2025 · similarity 0.88)
- Spectral Feedback for Test-Time Alignment of Protein Diffusion Models (NeurIPS 2026 · similarity 0.88)
- D-DOIT: Training-free Adaptation of Discrete Diffusion via Doob's h-Transform (NeurIPS 2026 · similarity 0.88)
- Diffusion on Language Model Encodings for Protein Sequence Generation (ICML 2025 · similarity 0.87)
- Derivative-Free Guidance in Continuous and Discrete Diffusion Models with Soft Value-based Decoding (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.
