ReDit: Reward Dithering for Improved LLM Policy Optimization

• URL: http://arxiv.org/abs/2506.18631v2
• Date: Tue, 24 Jun 2025 07:07:57 GMT
• Title: ReDit: Reward Dithering for Improved LLM Policy Optimization
• Abstract summary: DeepSeek-R1 has successfully enhanced Large Language Model (LLM) reasoning capabilities through its rule-based reward system.<n>While it's a ''perfect'' reward system that effectively mitigates reward hacking, such reward functions are often discrete.<n>We propose ReDit (Reward Dithering), a method that dithers the discrete reward signal by adding simple random noise.
• Score: 6.841631032347429
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: DeepSeek-R1 has successfully enhanced Large Language Model (LLM) reasoning capabilities through its rule-based reward system. While it's a ''perfect'' reward system that effectively mitigates reward hacking, such reward functions are often discrete. Our experimental observations suggest that discrete rewards can lead to gradient anomaly, unstable optimization, and slow convergence. To address this issue, we propose ReDit (Reward Dithering), a method that dithers the discrete reward signal by adding simple random noise. With this perturbed reward, exploratory gradients are continuously provided throughout the learning process, enabling smoother gradient updates and accelerating convergence. The injected noise also introduces stochasticity into flat reward regions, encouraging the model to explore novel policies and escape local optima. Experiments across diverse tasks demonstrate the effectiveness and efficiency of ReDit. On average, ReDit achieves performance comparable to vanilla GRPO with only approximately 10% the training steps, and furthermore, still exhibits a 4% performance improvement over vanilla GRPO when trained for a similar duration. Visualizations confirm significant mitigation of gradient issues with ReDit. Moreover, theoretical analyses are provided to further validate these advantages.

Related papers

• Tail-Likelihood Reinforcement Learning [63.90013981311359]
  Tail-Likelihood Reinforcement Learning (TailRL) maximizes the log-probability of exceeding a randomly chosen reward threshold.<n>TailRL requires only a simple modification to the advantage function, making it compatible with existing reinforcement learning pipelines.
  arXiv  Detail & Related papers  (2026-09-02T14:54:45Z)
• RSPO: Reward-Swap Policy Optimization for Multi-Turn LLM Agents [24.215311212134882]
  We propose Reward-Swap Policy Optimization (RSPO) to leverage the rich information from dense process rewards to facilitate training with outcome rewards.<n>RSPO ensures the diversity of sampled trajectories while guaranteeing consistency between the optimization objective and the true outcome rewards.
  arXiv  Detail & Related papers  (2026-07-06T06:32:39Z)
• MARBLE: Multi-Aspect Reward Balance for Diffusion RL [71.6241143519038]
  Reinforcement learning has become the dominant approach for aligning diffusion models with human preferences.<n>Existing practice deal with multiple rewards by training one specialist model per reward.<n>We propose MARBLE, a gradient-space optimization framework that maintains independent advantage estimators for each reward.
  arXiv  Detail & Related papers  (2026-05-07T16:20:42Z)
• Reward Sharpness-Aware Fine-Tuning for Diffusion Models [10.026496861838448]
  Reinforcement learning from human feedback has proven effective in aligning large language models with human preferences.<n>RDRL remains susceptible to reward hacking, where the reward score increases without corresponding improvements in perceptual quality.<n>We introduce methods that exploit gradients from a robustified reward model without requiring its retraining.
  arXiv  Detail & Related papers  (2026-03-22T11:32:50Z)
• IR$^3$: Contrastive Inverse Reinforcement Learning for Interpretable Detection and Mitigation of Reward Hacking [67.20568716300272]
  Reinforcement Learning from Human Feedback (RLHF) enables powerful LLM alignment but can introduce reward hacking.<n>We introduce IR3 (Interpretable Reward Reconstruction and Rectification), a framework that reverse-engineers, interprets, and surgically repairs the implicit objectives driving RLHF-tuned models.<n>We show that IR3 achieves 0.89 correlation with ground-truth rewards, identifies hacking features with over 90% precision, and significantly reduces hacking behaviors while maintaining capabilities within 3% of the original model.
  arXiv  Detail & Related papers  (2026-02-23T01:14:53Z)
• VeRPO: Verifiable Dense Reward Policy Optimization for Code Generation [43.206705536310245]
  We introduce textbfVeRPO (textbfVerifiable Dtextbfense textbfReward textbfPolicy textbfOptimization), a novel RL framework for code generation that synthesizes textitrobust and dense rewards fully grounded in verifiable execution feedback.<n>VeRPO consistently outperforms outcome-driven and RM-based baselines, achieving up to +8.83% gain in pass@1 with negligible time cost ( 0.02%) and zero
  arXiv  Detail & Related papers  (2026-01-07T02:29:49Z)
• 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)
• GARDO: Reinforcing Diffusion Models without Reward Hacking [54.841464430913476]
  Fine-tuning diffusion models via online reinforcement learning (RL) has shown great potential for enhancing text-to-image alignment.<n>The mismatch often leads to reward hacking, where proxy scores increase while real image quality deteriorates and generation diversity collapses.<n>We propose Gated and Adaptive Regularization with Diversity-aware Optimization (GARDO) to address the competing demands of sample efficiency, effective exploration, and mitigation of reward hacking.
  arXiv  Detail & Related papers  (2025-12-30T10:55:45Z)
• 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)
• G$^2$RPO: Granular GRPO for Precise Reward in Flow Models [74.21206048155669]
  We propose a novel Granular-GRPO (G$2$RPO) framework that achieves precise and comprehensive reward assessments of sampling directions.<n>We introduce a Multi-Granularity Advantage Integration module that aggregates advantages computed at multiple diffusion scales.<n>Our G$2$RPO significantly outperforms existing flow-based GRPO baselines.
  arXiv  Detail & Related papers  (2025-10-02T12:57:12Z)
• Intra-Trajectory Consistency for Reward Modeling [67.84522106537274]
  We develop an intra-trajectory consistency regularization to enforce that adjacent processes with higher next-token generation probability maintain more consistent rewards.<n>We show that the reward model trained with the proposed regularization induces better DPO-aligned policies and achieves better best-of-N (BON) inference-time verification results.
  arXiv  Detail & Related papers  (2025-06-10T12:59:14Z)
• A Minimalist Approach to LLM Reasoning: from Rejection Sampling to Reinforce [68.99924691391048]
  We revisit GRPO from a reinforce-like algorithm perspective and analyze its core components.<n>We find that a simple rejection sampling baseline, RAFT, yields competitive performance than GRPO and PPO.<n>Motivated by this insight, we propose Reinforce-Rej, a minimal extension of policy gradient that filters both entirely incorrect and entirely correct samples.
  arXiv  Detail & Related papers  (2025-04-15T16:15:02Z)
• Shaping Sparse Rewards in Reinforcement Learning: A Semi-supervised Approach [7.200081267352692]
  Experimental results in Atari and robotic manipulation demonstrate that our method outperforms supervised-based approaches in reward inference.<n>In more sparse-reward environments, our method achieves up to twice the peak scores compared to supervised baselines.
  arXiv  Detail & Related papers  (2025-01-31T13:35:19Z)
• Highly Efficient Self-Adaptive Reward Shaping for Reinforcement Learning [5.242869847419834]
  Reward shaping is a technique in reinforcement learning that addresses the sparse-reward problem by providing more frequent and informative rewards.<n>We introduce a self-adaptive and highly efficient reward shaping mechanism that incorporates success rates derived from historical experiences as shaped rewards.<n>Our method is validated on various tasks with extremely sparse rewards, demonstrating notable improvements in sample efficiency and convergence stability over relevant baselines.
  arXiv  Detail & Related papers  (2024-08-06T08:22:16Z)
• REBEL: Reward Regularization-Based Approach for Robotic Reinforcement Learning from Human Feedback [61.54791065013767]
  A misalignment between the reward function and human preferences can lead to catastrophic outcomes in the real world.<n>Recent methods aim to mitigate misalignment by learning reward functions from human preferences.<n>We propose a novel concept of reward regularization within the robotic RLHF framework.
  arXiv  Detail & Related papers  (2023-12-22T04:56:37Z)
• DreamSmooth: Improving Model-based Reinforcement Learning via Reward Smoothing [60.21269454707625]
  DreamSmooth learns to predict a temporally-smoothed reward, instead of the exact reward at the given timestep. We show that DreamSmooth achieves state-of-the-art performance on long-horizon sparse-reward tasks.
  arXiv  Detail & Related papers  (2023-11-02T17:57:38Z)
• Mind the Gap: Offline Policy Optimization for Imperfect Rewards [14.874900923808408]
  We propose a unified offline policy optimization approach, textitRGM (Reward Gap Minimization), which can handle diverse types of imperfect rewards. By exploiting the duality of the lower layer, we derive a tractable algorithm that enables sampled-based learning without any online interactions.
  arXiv  Detail & Related papers  (2023-02-03T11:39:50Z)
• Maximum-Likelihood Inverse Reinforcement Learning with Finite-Time Guarantees [56.848265937921354]
  Inverse reinforcement learning (IRL) aims to recover the reward function and the associated optimal policy. Many algorithms for IRL have an inherently nested structure. We develop a novel single-loop algorithm for IRL that does not compromise reward estimation accuracy.
  arXiv  Detail & Related papers  (2022-10-04T17:13:45Z)
• Policy Gradient Bayesian Robust Optimization for Imitation Learning [49.881386773269746]
  We derive a novel policy gradient-style robust optimization approach, PG-BROIL, to balance expected performance and risk. Results suggest PG-BROIL can produce a family of behaviors ranging from risk-neutral to risk-averse.
  arXiv  Detail & Related papers  (2021-06-11T16:49:15Z)
• Model-free Policy Learning with Reward Gradients [9.847875182113137]
  We develop the textitReward Policy Gradient estimator, a novel approach that integrates reward gradients without learning a model. Our method also boosts the performance of Proximal Policy Optimization on different MuJoCo control tasks.
  arXiv  Detail & Related papers  (2021-03-09T00:14:13Z)
• Self-Supervised Online Reward Shaping in Sparse-Reward Environments [36.01839934355542]
  We propose a novel reinforcement learning framework that performs self-supervised online reward shaping. The proposed framework alternates between updating a policy and inferring a reward function. Experimental results on several sparse-reward environments demonstrate that the proposed algorithm is significantly more sample efficient than the state-of-the-art baseline.
  arXiv  Detail & Related papers  (2021-03-08T03:28:04Z)
• Extrapolation for Large-batch Training in Deep Learning [72.61259487233214]
  We show that a host of variations can be covered in a unified framework that we propose. We prove the convergence of this novel scheme and rigorously evaluate its empirical performance on ResNet, LSTM, and Transformer.
  arXiv  Detail & Related papers  (2020-06-10T08:22:41Z)

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.