Diffusion Reinforcement Learning via Centered Reward Distillation

• URL: http://arxiv.org/abs/2603.14128v1
• Date: Sat, 14 Mar 2026 21:29:33 GMT
• Title: Diffusion Reinforcement Learning via Centered Reward Distillation
• Abstract summary: We present textbf Reward Distillation (CRD), a diffusion RL framework derived from KL-regularized reward models built on forward-process fine-tuning.<n>To enable reliable text-to-image fine-tuning, we introduce techniques that explicitly control distribution drift.<n>Experiments on text-to-image post-training with textttGenEval and textttOCR rewards show that competitive SOTA reward optimization results with fast convergence and reduced reward hacking preference.
• Score: 35.979608265594685
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Diffusion and flow models achieve State-Of-The-Art (SOTA) generative performance, yet many practically important behaviors such as fine-grained prompt fidelity, compositional correctness, and text rendering are weakly specified by score or flow matching pretraining objectives. Reinforcement Learning (RL) fine-tuning with external, black-box rewards is a natural remedy, but diffusion RL is often brittle. Trajectory-based methods incur high memory cost and high-variance gradient estimates; forward-process approaches converge faster but can suffer from distribution drift, and hence reward hacking. In this work, we present \textbf{Centered Reward Distillation (CRD)}, a diffusion RL framework derived from KL-regularized reward maximization built on forward-process-based fine-tuning. The key insight is that the intractable normalizing constant cancels under \emph{within-prompt centering}, yielding a well-posed reward-matching objective. To enable reliable text-to-image fine-tuning, we introduce techniques that explicitly control distribution drift: (\textit{i}) decoupling the sampler from the moving reference to prevent ratio-signal collapse, (\textit{ii}) KL anchoring to a CFG-guided pretrained model to control long-run drift and align with the inference-time semantics of the pre-trained model, and (\textit{iii}) reward-adaptive KL strength to accelerate early learning under large KL regularization while reducing late-stage exploitation of reward-model loopholes. Experiments on text-to-image post-training with \texttt{GenEval} and \texttt{OCR} rewards show that CRD achieves competitive SOTA reward optimization results with fast convergence and reduced reward hacking, as validated on unseen preference metrics.

Related papers

• Scaling Reinforcement Learning for Diffusion Models via Velocity Matching [66.34541945134916]
  RewardBench fine-tuning is an important tool for adapting diffusion models to human preferences and task-specific objectives.<n>Unlike autoregressive models, diffusion models do not provide tractable likelihoods for generated samples.<n>We propose reward-based velocity matching (RVM), a simple trajectory-free update that acts directly on the velocity field.
  arXiv  Detail & Related papers  (2026-08-24T17:13:43Z)
• Explicit Critic Guidance for Aligning Diffusion Models [27.6704888666835]
  We propose a state-aligned latent actor-critic framework for diffusion post-training.<n>Our method consistently outperforms prior group-relative RL and actor-critic baselines on single-reward and multi-reward benchmarks.
  arXiv  Detail & Related papers  (2026-05-26T22:20:51Z)
• Clipping Bottleneck: Stabilizing RLVR via Stochastic Recovery of Near-Boundary Signals [83.0127582612634]
  Near-boundary Rescue (NSR) is a minimal, plug-and-play modification that retains out-of-bound tokens to recover lost signals.<n>NSR substantially improves training stability and delivers consistent gains over strong baselines such as DAPO and GSPO.
  arXiv  Detail & Related papers  (2026-05-21T16:45:31Z)
• Diffusion Controller: Framework, Algorithms and Parameterization [54.82539154511621]
  We present a unified control-theoretic view that casts reverse diffusion sampling as state-only control within (generalized) linearly-solvable Markov Decision Processes.<n>Under this framework, control acts by reweighting the pretrained reverse-time transition kernels, balancing terminal objectives against an $f$-divergence cost.<n>Experiments on Stable Diffusion v1.4 show consistent gains in preference-alignment win rates and improved quality-efficiency trade-offs.
  arXiv  Detail & Related papers  (2026-03-07T01:49:59Z)
• DeRaDiff: Denoising Time Realignment of Diffusion Models [45.23864679259705]
  How does one choose the right regularization strength?<n>DeRaDiff modulates the regularization strength during sampling to emulate models trained at other regularization strengths without any additional training or finetuning.<n>Our method consistently provides a strong approximation for models aligned entirely from scratch at different regularization strengths.
  arXiv  Detail & Related papers  (2026-01-28T02:53:39Z)
• GDRO: Group-level Reward Post-training Suitable for Diffusion Models [55.948229011478304]
  Group-level rewards successfully align the model with the targeted reward.<n>Group-level Direct Reward Optimization (GDRO) is a new post-training paradigm for group-level reward alignment.<n>GDRO supports full offline training that saves the large time cost for image rollout sampling.<n>It is diffusion-sampler-independent, which eliminates the need for the ODE-to-SDE approximation to obtainity.
  arXiv  Detail & Related papers  (2026-01-05T11:47:18Z)
• Data-regularized Reinforcement Learning for Diffusion Models at Scale [99.01056178660538]
  We introduce Data-regularized Diffusion Reinforcement Learning ( DDRL), a novel framework that uses the forward KL divergence to anchor the policy to an off-policy data distribution.<n>With over a million GPU hours of experiments and ten thousand double-blind evaluations, we demonstrate that DDRL significantly improves rewards while alleviating the reward hacking seen in RLs.
  arXiv  Detail & Related papers  (2025-12-03T23:45:07Z)
• Stabilizing Reinforcement Learning with LLMs: Formulation and Practices [61.361819972410046]
  We show why and under what conditions the true sequence-level reward can be optimized via a surrogate token-level objective in policy gradient methods such as REINFORCE.<n>This insight provides a principled explanation for the crucial role of several widely adopted techniques in stabilizing RL training.
  arXiv  Detail & Related papers  (2025-12-01T07:45:39Z)
• VARD: Efficient and Dense Fine-Tuning for Diffusion Models with Value-based RL [28.95582264086289]
  VAlue-based Reinforced Diffusion (VARD) is a novel approach that first learns a value function predicting expection of rewards from intermediate states.<n>Our method maintains proximity to the pretrained model while enabling effective and stable training via backpropagation.
  arXiv  Detail & Related papers  (2025-05-21T17:44:37Z)
• Efficient Diffusion as Low Light Enhancer [63.789138528062225]
  Reflectance-Aware Trajectory Refinement (RATR) is a simple yet effective module to refine the teacher trajectory using the reflectance component of images. textbfReflectance-aware textbfDiffusion with textbfDistilled textbfTrajectory (textbfReDDiT) is an efficient and flexible distillation framework tailored for Low-Light Image Enhancement (LLIE)
  arXiv  Detail & Related papers  (2024-10-16T08:07:18Z)
• DR-Tune: Improving Fine-tuning of Pretrained Visual Models by Distribution Regularization with Semantic Calibration [38.4461170690033]
  We propose a novel fine-tuning framework, namely distribution regularization with semantic calibration (DR-Tune) DR-Tune employs distribution regularization by enforcing the downstream task head to decrease its classification error on the pretrained feature distribution. To alleviate the interference by semantic drift, we develop the semantic calibration (SC) module.
  arXiv  Detail & Related papers  (2023-08-23T10:59:20Z)
• Exploiting Diffusion Prior for Real-World Image Super-Resolution [75.5898357277047]
  We present a novel approach to leverage prior knowledge encapsulated in pre-trained text-to-image diffusion models for blind super-resolution. By employing our time-aware encoder, we can achieve promising restoration results without altering the pre-trained synthesis model.
  arXiv  Detail & Related papers  (2023-05-11T17:55:25Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.