The paper introduces DPRM, a plug‑in token‑ordering module for discrete diffusion models that uses a Doob h‑transform to convert terminal rewards into per‑position process rewards. By estimating these rewards from generation progress, confidence, and optional state, DPRM reorders tokens without altering the underlying model or sampler. Experiments on nine open‑source hosts show significant gains in reasoning, numeric VQA, visual‑codebook ordering, and preference‑conditioned generation, with improvements ranging from 8.97 to 53.3 points across diverse tasks.
arXiv:2604. 24357v2 Announce Type: replace-cross Abstract: Diffusion language models generate without a fixed left-to-right order, leaving token ordering as a central algorithmic choice.
arXiv:2609.33803v2 Announce Type: replace-cross Abstract: Reward models underpin the alignment of large language models, yet the dominant designs reduce each prompt--response pair to a point estimate...
arXiv:2608.21501v1 Announce Type: new Abstract: Credit assignment in large-language-model reinforcement learning (LLM RL) can be separated into three objects: evidence about success, a transport oper...
VGAS: Variance-Reduced Guidance and Adaptive Selection for Training-Free Reward Alignment in Discrete Diffusion proposes a new inference-time framework that improves steering of frozen masked discrete diffusion models. By reducing the variance of guidance estimates, applying reward tilting to clean-token logits, and adapting the selection temperature at each step, VGAS addresses three default choices in existing pipelines. Experiments on regulatory DNA, protein, and small-molecule benchmarks show that VGAS achieves the best training-free reward performance and matches or surpasses reward-fine-tuned generators.
arXiv:2608. 03929v1 Announce Type: new Abstract: Aligning diffusion models with human preferences usually relies on a sparse terminal reward evaluated on the final generated samples, presenting a severe temporal credit-assignment challenge across the multi-step denoising process.
arXiv:2607. 25136v1 Announce Type: new Abstract: Research on preference optimization often varies the training objective while holding the data fixed.
The paper introduces committed reveal sampling (CRS), a training‑free sampler for uniform discrete diffusion models that stores selected argmax tokens as persistent context for subsequent predictions. CRS keeps these tokens visible in later model inputs, which theoretically prevents Bayes error from increasing as noise decreases and encourages consistent sequence‑level choices. Empirical tests on Duo‑distilled data show that CRS without top‑p truncation achieves lower generative perplexity than fixed‑p baselines across various numbers of function evaluations, offering a more favorable perplexity–entropy trade‑off.
Informed Masking (IM) is a new technique for aligning Diffusion Large Language Models (dLLMs) with Reinforcement Learning (RL). It identifies a systematic upstream/downstream token structure in dLLM rollouts and shows that masking downstream tokens creates better subproblems for likelihood estimation. When integrated into three state‑of‑the‑art dLLM RL methods on LLaDA‑8B‑Instruct, IM yields up to 2.01%, 8.68%, and 5.77% relative average gains on math and planning benchmarks while improving training stability.
arXiv:2606. 19162v1 Announce Type: new Abstract: Score- and flow-matching models often rely on preference-based reinforcement learning for two purposes: aligning with subjective preferences and, surprisingly, recovering properties such as visual realism and coherent object structure that matching-based training is intended to learn from the data itself.
arXiv:2606. 08480v1 Announce Type: cross Abstract: Reinforcement learning (RL) presents a promising avenue for enhancing generative recommendation beyond supervised imitation, leveraging reward signals to guide policy improvement.
arXiv:2607. 01170v1 Announce Type: cross Abstract: Generative reasoning re-rankers achieve strong recommendation accuracy by emitting a chain-of-thought before re-ordering a candidate list, but they are slow at inference: an autoregressive (AR) decoder spends one sequential forward pass per reasoning token, and the reasoning trace far exceeds the ranking it produces.
arXiv:2608.30427v1 Announce Type: cross Abstract: Speculative decoding speeds up generation with an efficient draft model (drafter) that proposes tokens for a target model to verify in one pass, pres...