5arXiv cs.LG (Machine Learning)·
Selective timestep weighting and advantage-based replay improve diffusion RLHF sample efficiency by up to 6×
A new arXiv preprint proposes two complementary techniques to improve feedback efficiency in diffusion model RLHF: a per-timestep weighting scheme grounded in PPO convergence theory, and a replay mechanism that prioritizes informative trajectories to reduce redundant reward queries. Together, the methods achieve up to 6× improvement in sample efficiency over standard diffusion RLHF baselines under identical hyperparameter settings. The work addresses a practical bottleneck—feedback cost—that limits real-world deployment of RLHF-aligned diffusion models.
Related guides (3)
Related events (8)
Finetune Stable Diffusion Models with DDPO via TRL
Hugging Face's TRL library adds support for DDPO (Denoising Diffusion Policy Optimization), enabling reinforcement learning-based finetuning of Stable Diffusion models. This extends TRL's RLHF tooling beyond language models to image generation, allowing reward-driven optimization of diffusion models. The post demonstrates practical usage of the new DDPO trainer within the TRL ecosystem.
Kolmogorov Regression lifts diffusion policies to Cameron-Martin space for robust long-horizon control
Researchers introduce a backward Kolmogorov equation framework that reformulates diffusion policy training as a deterministic boundary-value PDE problem in Cameron-Martin space, replacing stochastic score matching. The approach uses a precision-weighted Cameron-Martin loss and a Kolmogorov residual as an inference-time failure detector, yielding convergence guarantees tied to kernel effective rank rather than action dimension. Validation on the PushT manipulation benchmark shows 17% improvement in episode reward and 67.6% reduction in inter-step drift; a 6-station manufacturing scheduling task shows 28.4% lower RMSE than LSTM baselines and 96% reduction in deadlock events via Hamilton-Jacobi reachability certification.
Using LoRA for Efficient Stable Diffusion Fine-Tuning
This Hugging Face blog post explains how Low-Rank Adaptation (LoRA) can be applied to fine-tune Stable Diffusion models efficiently. LoRA reduces the number of trainable parameters by decomposing weight updates into low-rank matrices, enabling fine-tuning on consumer hardware with significantly less memory. The post covers practical implementation details using the diffusers library.
RELEX: Extrapolating LLM RLVR Training via Rank-1 Parameter Trajectories
This paper demonstrates that RLVR weight update trajectories are extremely low-rank and near-linearly predictable, with a rank-1 approximation capturing most downstream performance gains. The authors propose RELEX, a compute-efficient method that observes a short training window, estimates the rank-1 subspace, and extrapolates future checkpoints via linear regression—requiring no additional training. Evaluated on Qwen2.5-Math-1.5B, Qwen3-4B-Base, and Qwen3-8B-Base, RELEX matches or exceeds full RLVR performance using as few as 15% of training steps, and can extrapolate up to 10–20× beyond the observed prefix. The authors attribute the method's effectiveness to a denoising effect from rank-1 projection that discards stochastic optimization noise.
ACPO: Adaptive Clip Policy Optimization improves RLVR training for LLM reasoning
A new arXiv preprint provides theoretical analysis of Reinforcement Learning from Verifiable Rewards (RLVR) updates, identifying off-policy degree and gradient expectation as key factors governing update dynamics. The authors show that differences in gradient steps per rollout substantially affect importance sampling ratio distributions and which tokens dominate updates. Based on this analysis, they propose Adaptive Clip Policy Optimization (ACPO), which adjusts clipping boundaries per token group by empirical variance of importance sampling ratios, outperforming DAPO and CISPO baselines on 3B and 7B models across math, tabular QA, and logic benchmarks.
DRPO: Smooth divergence regularization replaces hard masking in LLM RL training
A new arXiv preprint proposes Divergence Regularized Policy Optimization (DRPO), a method that replaces the hard trust-region mask used in DPPO with a smooth advantage-weighted quadratic regularizer on policy shift. The approach addresses a known weakness in PPO and GRPO where importance ratios poorly proxy distributional shift in long-tailed vocabularies, and in DPPO where gradient signals are discarded rather than corrected at trust-region boundaries. Experiments across model scales, architectures, and precision settings show improved stability and efficiency in LLM RL post-training.
DRIFT: Decoupled Rollouts and Importance-Weighted Fine-Tuning for Efficient Multi-Turn Optimization
DRIFT is a training framework that bridges online RL and offline SFT for multi-turn LLM optimization by exploiting the theoretical equivalence between KL-regularized RL and importance-weighted supervised learning. It decouples rollout generation from policy optimization: trajectories are sampled from a fixed reference policy offline, weighted by return-based importance scores, and used for weighted SFT. Empirically, DRIFT matches or exceeds multi-turn RL baselines while retaining the efficiency and simplicity of standard supervised fine-tuning. Code is publicly released.
SURGE: Approximation-free Training-Free Particle Filter for Diffusion Surrogate
The paper introduces URGE (Unbiased Resampling via Girsanov Estimation), a derivative-free inference-time scaling algorithm for diffusion models that performs path-wise importance reweighting using a Girsanov change of measure. Unlike existing inference-time guidance methods, URGE requires no score, Hessian, or PDE evaluations, attaching multiplicative weights to simulated trajectories and periodically resampling. The authors establish a theoretical equivalence between path-wise and particle-wise sequential Monte Carlo (SMC), guaranteeing unbiased terminal distributions. Empirically, URGE outperforms existing inference-time guidance baselines on synthetic tests and diffusion-model benchmarks while being simpler to implement.