AI Paper Atlas

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

ICLR 2026 · Reinforcement learning › RL from feedback · approach: Reward modeling versus direct optimization · 10 citations (OpenAlex)

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 changedWhereMeasuredEffect
increase samples per prompt N from 2 (pairwise) to 25 (ablation)SDPO · Protein inverse foldingPred-ddG: 0.529 -> 1.119Improved
increase N from 25 to 100 (ablation)SDPO · Protein inverse foldingPred-ddG: 1.119 -> 1.061No clear effect
lower beta (weaker KL regularization) (ablation)SDPO · DNA design / protein foldingPred-Activity / Pred-ddGMixed
add iterative labeling (2 rounds) (ablation)SDPO · DNA designPred-ActivityImproved
SDPO vs DRAKESDNA design · DNA enhancer design (HepG2)Pred-Activity: 5.61 -> 6.30Improved
SDPO vs diffu-GRPOprotein inverse folding · Protein inverse foldingSuccess rate (%): 37.2 -> 75.5Improved
SDPO vs DRAKESprotein inverse folding · Protein inverse foldingSuccess rate (%): 78.6 -> 75.5Hurt
apply SDPO to LLaDA-8B-InstructLLaDA-8B-Instruct · GSM8Kaccuracy: 78.6 -> 81.2Improved
D2-DPO on LLaDA-8B-InstructLLaDA-8B-Instruct · GSM8Kaccuracy: 78.6 -> 78.1No clear effect
SDPO vs on-policy training costDRAKES · DNA designseconds per step: 6.02 -> 0.77Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy