Reinforcement_Learning
-
Health-SCORE: Towards Scalable Rubrics for Improving Health-LLMs
2026-10-25 01:24:23Zhichao Yang, Sepehr Janghorbani, Dongxu Zhang, Jun Han, Qian Qian, Andrew Ressler II, Gregory D. Lyng, Sanjit Singh Batra, Robert E. TillmanAbstract
Rubrics are essential for evaluating open-ended LLM responses, especially in safety-critical domains such as healthcare. However, creating high-quality and domain-specific rubrics typically requires significant human expertise time and development cost, making rubric-based evaluation and training difficult to scale. In this work, we introduce Health-SCORE, a generalizable and scalable rubric-based training and evaluation framework that substantially reduces rubric development costs without sacrificing performance. We show that Health-SCORE provides two practical benefits beyond standalone evaluation: it can be used as a structured reward signal to guide reinforcement learning with safety-aware supervision, and it can be incorporated directly into prompts to improve response quality through in-context learning. Across open-ended healthcare tasks, Health-SCORE achieves evaluation quality comparable to human-created rubrics while significantly lowering development effort, making rubric-based evaluation and training more scalable.
Abstract (translated)
URL
https://arxiv.org/abs/2601.18706
PDF
-
Curriculum-Driven Deep Reinforcement Learning for Autonomous Far-Range Rocket Landing
2026-10-24 19:57:44Atin Chowdhury, Piyush JoshiAbstract
Autonomous vertical landing of reusable launch vehicles is a critical yet formidable challenge in aerospace engineering, primarily due to highly nonlinear dynamics, stringent precision require- ments, and inherent environmental uncertainties. While reinforcement learning (RL) has emerged as a promising alternative to traditional model-based control, existing approaches frequently rely on flat training paradigms and highly simplified initial conditions, drastically limiting their real-world applicability. This paper proposes a comprehensive curriculum-driven deep reinforcement learning framework designed to master both far-range navigation and precision landing. By employing a three-stage curriculum augmented with a critic warm-up procedure and a rehearsal mechanism to prevent catastrophic forgetting, alongside a high-dimensional observation vector utilizing a full 3 X 3 rotation-matrix, the agent is trained to navigate from an initial horizontal distance of 1,000 to 1,500 units vastly farther than prior benchmarks before executing a soft touchdown. Crucially, we intro- duce a redesigned reward function that reduces unnecessary computational overhead and addresses limitations of earlier formulations, including poorly calibrated reward components and potentially conflicting optimization objectives. The redesigned function bounds the per-step reward magnitude by (O(1)), while its progress term accumulates the total distance closed over the trajectory. The available 100,000-episode results report a 99.98% success rate for the proposed policy and 68.31% for the baseline under the evaluation conditions. In the evaluation, the proposed policy retained 69.1% of its initial fuel capacity and achieved a mean final distance from the pad centre of 7.07 units, compared with 13.10 units for the baseline. Landing visualizations are available in Video 1 and Video 2.
Abstract (translated)
URL
https://arxiv.org/abs/2610.35954
PDF
-
Intervention-Certified Amplification: A Validity Gate for $sigma$-Weighted Post-Training
2026-10-24 19:49:59Leiyue Chen, Meng Wei, Xuefeng Hu, Wei LiuAbstract
Hybrid post-training often mixes supervised fine-tuning (SFT) and reinforcement learning (RL), using training-side signals to set each sample's balance. Such signals characterize the anchor data and policy, but not whether the emphasized behavior helps in the target task and evaluation setting. This mismatch can reactivate and amplify pretrained behavior even after its literal form is removed from supervision. We demonstrate this with DIAA (Dual-Implicit Adversarial Amplification), a per-sample $\sigma$-weighted SFT/RL hybrid. With all literal Python fences removed from its anchor, DIAA emits unexecuted Tool-Integrated Reasoning (TIR) on $29%$ of MATH-500 prompts, three times vanilla GRPO. Paired interventions show $10.6$ percentage points of accuracy recovery when TIR is suppressed and comparable contributions from SFT imitation and $\sigma$-amplification; a student trained on a $0%$-TIR teacher anchor emits it at $45.8%$. Intervention-Certified Amplification (ICA) gates $\sigma$ with paired-rollout target evidence. Frozen and online ICA reduce TIR to $13.8%$ and $11.2%$, halve residual drag, improve accuracy from $47.5%$ to $48.7%$ and $49.3%$, and transfer to held-out GSM8K without retuning.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33882
PDF
-
RoboLongR: Causally Consistent Rewards Teach Robots to Progressively Master Long-Horizon Manipulation
2026-10-24 19:07:21Zixuan Chen, Lin Cao, Fenggang Zhang, Haotong Zhang, Yangtao Chen, Jieqi Shi, Jing Huo, Yang GaoAbstract
Long-horizon manipulation requires actions whose validity depends on earlier interactions, yet visual progress can conceal missing task prerequisites. We propose RoboLongR, an offline-to-online framework that learns history-dependent rewards to guide policy adaptation to new tasks. Its central objective is causal consistency: aligning rewarded progress with prerequisites supported by execution history. Stage-shuffle supervision encourages execution-order sensitivity, recurrent memory preserves evidence of earlier interactions, and two-round training refines trajectory segments using predicted progress. The resulting rewards guide stage-conditioned policy refinement from behavior cloning to online reinforcement learning. Sparse terminal feedback connects policy experience to reward learning by triggering correction for repeated high-scoring failures. Pretrained on demonstrations from 88 RLBench tasks and Open-X data, RoboLongR achieves 31.8% average success on 18 long-horizon tasks excluded from reward pretraining, exceeding the strongest evaluated baseline by 9.2 percentage points. Across four real-robot tasks, it achieves 57.5% success, a 32.5-percentage-point improvement over the comparison method. Reward-level controls probe sensitivity to synthetic order perturbations, while ablations examine the contributions of individual components to policy improvement.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33863
PDF
-
EdgeDeID: A 0.6B Clinical De-identifier Trained Without Real Patient Data
2026-10-24 18:20:52Sina Daneshgar, Raya Namazi, Amirali Khosravi, Pouria Rouzrokh, Bardia KhosraviAbstract
De-identifying clinical notes with a hosted language model discloses the information the step is meant to protect. We present EdgeDeID, a 0.6B-parameter decoder-only de-identifier that runs locally and is trained without real patient text: three LLM teachers write synthetic notes, a single labeler re-annotates them into one convention, and the model learns to emit a typed XML schema of the identifiers it finds. On a 100-note synthetic in-domain benchmark with manually verified annotations it reaches 0.994 micro-F1 and leaves no missed identifier in 99% of notes, against 0.857 and 12% for the strongest of eight off-the-shelf systems. Quantized, it runs on CPU in 615 MB. Of three additions to supervised training, only a verifier adapter helps: it separates complete from incomplete outputs of other systems at 0.94 AUROC, whereas reinforcement learning with a recall-weighted reward and a rule-based post-processing layer leave in-domain results unchanged. On MIMIC ICU nursing notes F1 falls to 0.48. The errors follow two properties our synthetic notes lack, telegraphic style and low identifier density, and three hand-written in-context examples recover 0.08 F1 without annotated target data.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33839
PDF
-
EnvLunch: The Environment as a Free Lunch for Learning to Interact
2026-10-24 17:29:24Wei Liu, Min Gao, Yali Du, Yulan HeAbstract
For agentic data analysis, language-model agents often require a large number of annotated tasks to learn how to interact effectively with executable environments (i.e., databases). While high-quality annotated tasks require heavy manual effort, an executable environment already provides rich interaction structure and verifiable outcomes. In this sense, the environment itself offers a free lunch, an additional set of verifiable tasks that require no human annotation. We propose EnvLunch, a warm-up stage that turns this environmental signal into agent training without human-written or model-generated task descriptions, before downstream agentic reinforcement learning. EnvLunch samples valid action trajectories, executes them to obtain final observations, and trains an agent to recover a functionally equivalent trajectory given only its final observation. This label-free warm-up enables the model to outperform the baseline by over 9 points on average across all downstream training budgets, and with few labels, the baseline needs 8 to 32 times as many annotated examples to match the accuracy that the EnvLunch model reaches at its smallest training budget. Letting the environment alone define the task yields a larger gain than having the model write its own tasks over the same environments, and this warm-up also improves fine-tuning on benchmarks beyond data analysis. EnvLunch therefore saves annotated tasks and provides a better starting point for fine-tuning agentic models when such tasks are limited.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33817
PDF
-
The Hitchhiker's Guide to System 1 Models: Fast Decision-Making Beyond LLMS
2026-10-24 16:26:01Haggai RoitmanAbstract
Large Language Models excel at reasoning and generation but their autoregressive, token-by-token inference makes them unnecessarily slow, expensive, and poorly calibrated for the many routine decisions (routing, gating, scoring, classification) that modern AI systems require at sub-second latency. A growing class of models addresses this gap: System 1 decision models, architectures that map an input directly to a typed, structured decision in a single forward pass. This guide provides a comprehensive, practitioner-oriented treatment of the field. We propose a five-family taxonomy covering encoder-based classifiers, reward models, non-autoregressive decision engines, contrastive decision models, and extraction models, and we provide in-depth coverage of each family's architecture, training methodology, and deployment trade-offs. A central theme is calibration: we develop the theory of proper scoring rules, analyze Reinforcement Learning for Calibrated Decisions (RLCD), and examine recently discovered phenomena including IIA violations, ordinal-scale bias, and few-shot recalibration under distribution shift. On the integration side, we describe how System 1 models serve as decision infrastructure within agentic workflows and System 1 + System 2 cascades that route between fast and slow models based on calibrated confidence. The guide covers 18 application domains, maps the rapidly growing open-source ecosystem (over 70 model reproductions in weeks), provides runnable code throughout, and includes a practitioner's checklist, an open-source model zoo, and 22 self-assessment questions with detailed answers. Spanning 190 references, it is intended as a unified entry point for researchers and engineers working at the intersection of fast decision-making, calibration, and production AI systems.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33775
PDF
-
Computer Vision Navigation for an Autonomous Vehicle Model Using Reinforcement Learning
2026-10-24 15:52:59Dinis Fernandes, Tiago Ribeiro, Ant\'onio Ribeiro, Carolina Lopes, Gil Lopes, A. Fernando RibeiroAbstract
This work aims to assess the capabilities of the reinforcement learning algorithm, Proximal Policy Optimization (PPO), in developing a policy capable of controlling the speed and steering angle of a vehicle model to navigate within simulated road lines. The control system's input is a camera, and its outputs are the rotational speed of the rear wheels and the steering angle of the front wheels, controlling the vehicle's speed and direction, respectively. The algorithm is trained and tested in a simulated environment, where it receives image frames from the vehicle's camera, processes them through a feature extraction algorithm, and feeds them into the policy. For each frame, the policy outputs speed and direction commands for the vehicle. The policy is iteratively updated to enhance its performance. The reinforcement learning algorithm successfully generates a policy capable of navigating the vehicle in a simplified environment. Potential improvements are suggested to enhance the system's robustness in more complex scenarios. Creating a simulation environment suitable for testing an advanced reinforcement learning algorithm is a complex task. This project offers a framework for other developers to test different approaches and algorithms. The tested algorithm, PPO, uses proximity intervals - an innovative approach in reinforcement learning. In some scenarios, the algorithm achieved better performance than predecessor algorithms, which are typically harder to implement and require tuning more hyperparameters.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33755
PDF
-
A Distribution-Matching Perspective on LLM Training Algorithms
2026-10-24 15:30:39Yanxing Huang, Bingzhang Yan, Zihan Tang, Wenzhe Li, Peng Li, Yang LiuAbstract
In this work, we propose a unified framework for training algorithms across different stages of modern large language model development, based on the perspective of distribution matching. We show that a broad class of language model training algorithms can be interpreted as minimizing a certain divergence between the current model and a target distribution, and that characterizing the properties of this target distribution provides substantial insight into the behavior of the corresponding training algorithms. By analyzing specific forms of the target distribution, we establish that On-Policy Distillation (OPD) enjoys an advantage over classical Knowledge Distillation (KD) when the current model distribution is close to the target distribution, while exhibiting a disadvantage when the two distributions are far apart. Furthermore, when the target distribution itself depends on the model being optimized, we show that the accumulation of estimation variance can induce a deterministic degradation of the target distribution. These results provide a theoretical explanation, at least in part, for the long-term behavior of existing Reinforcement Learning (RL) algorithms. With this intuition we further designed Conditional KL (CKL) regularization for mitigating the degradation of the target distribution. CKL can effectively preserve mode diversity while relaxing the constraint imposed by classical KL regularization on reward improvements. We believe this theoretical perspective can bring some new understanding of language model training algorithms and provide useful guidance for future research.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33740
PDF
-
Human-Feedback-Guided Multi-Agent Inverse Reinforcement Learning for Interactive Driving Behavior Modeling
2026-10-24 15:14:41Longhao Yan, Jinhao Liang, Kaidi YangAbstract
Modeling human driving behavior is fundamental to understanding driver decision-making and supporting autonomous vehicles (AVs) in mixed traffic. Reward learning offers a principled way to infer the latent objectives behind driving behaviors, yet most existing methods are formulated in single-agent settings and treat surrounding vehicles as dynamic obstacles. Directly extending reward learning to multi-agent scenarios is difficult because the joint action space grows exponentially with the number of interacting vehicles. In addition, existing approaches mainly rely on physical demonstrations and rarely exploit human feedback during reward recovery. To address these limitations, this paper proposes a Human-Feedback-Guided Multi-Agent Inverse Reinforcement Learning (HFG-MAIRL) framework for interactive driving behavior modeling. First, we develop a multi-agent maximum-entropy inverse reinforcement learning module that explicitly models each vehicle and its directly interacting neighbors in a localized Markov game, while using aggregated traffic states to capture the influence of the broader traffic environment. The resulting coupled multi-vehicle responses are approximated through finite marginal-response iterations over factorized vehicle-level policies, avoiding explicit enumeration of the full joint action space. Second, we incorporate human feedback as ordinal preference supervision through a Bradley-Terry likelihood. In our experiments, driver emotion labels are used as one instantiation of such feedback and converted into pairwise ordinal preference constraints. Experiments on the AIDE dataset show that HFG-MAIRL outperforms representative IRL baselines in decision modeling and produces interpretable feedback-aligned reward patterns.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33726
PDF
-
SWE-Ladder: Learning to Localize, Reproduce, and Patch via Progressive Privileged Self-Distillation
2026-10-24 13:27:18Xudong Cai, Zhengxi Lu, Yuchen Yan, Jian Shao, Jun Xiao, Yueting Zhuang, Yongliang ShenAbstract
Resolving a repository-level issue typically requires a software-engineering (SWE) agent to localize the faulty code, reproduce the failure, and patch the code. Reinforcement learning trains such agents from a single test result on the final patch, while on-policy self-distillation (OPSD) adds token-level supervision from a teacher with privileged evidence, exposing the full reference patch at every step. Neither provides stage-specific guidance for localization or reproduction. We propose SWE-Ladder, which turns the verified repair of each training instance into stage-specific evidence: the changed files and lines for localization, the fail-to-pass tests for reproduction, and the diff for patching. A rule-based router assigns each step of an end-to-end rollout to a stage based on its tool call, a frozen teacher sees only the matching evidence, and the stages are distilled in repair order while GRPO optimizes the full trajectory. With Qwen3-4B-Instruct-2507 on SWE-bench Verified under the R2E-Gym scaffold, SWE-Ladder improves pass@1 from 9.6% to 16.4%, compared with 10.4% for GRPO and 12.4% for OPSD with the full patch. Under the OpenHands scaffold, it improves pass@1 from 13.0% to 17.6%. Reversing the stage order under R2E-Gym reduces pass@1 to 11.4%.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33667
PDF
-
Physics-Informed Graph-Based Safe RL with Constraint-Specific Critics for Community Battery Coordination in Distribution Network
2026-10-24 13:11:29Chenghao Huang, Yue Yang, Hao WangAbstract
High penetrations of distributed energy resources (DERs) and fast-varying electric vehicle (EV) charging intensify voltage and line-limit violations in distribution networks (DNs), making safe, low-cost operation increasingly challenging. This paper proposes a physics-informed safe deep reinforcement learning (DRL) framework for coordinating community battery energy storage systems to mitigate constraint violations while minimizing operating cost. To represent nonlocal spatiotemporal interactions, we develop a dual-level physics-informed graph representation in which line impedance and capacity shape spatial information propagation, while power-balance, voltage-profile, and flow-reconstruction losses regularize the latent state toward control-relevant physical consistency. Building on this representation, we develop Lagrangian Dedicated-Critic PPO (LDCPPO), which implements constraint-specific value decomposition: independent reward and cost critics preserve heterogeneous voltage- and line-loading-related learning signals and combine their advantages only during the Lagrangian policy update. Case studies on DNs with high rooftop photovoltaic penetration and EV charging demonstrate that the proposed method reduces the key voltage-violation degree by up to 80.5% compared with the tested baselines, while also achieving the lowest voltage-violation count and line-overload degree with competitive operating cost.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33654
PDF
-
$xi$-DPO: Direct Preference Optimization via Ratio Reward Margin
2026-10-24 12:03:18Zhengyuan Fan, Zhonghua Wu, Yuxuan Du, Qun ChenAbstract
Reference-free preference optimization has emerged as an efficient alternative to reinforcement learning from human feedback, with Simple Preference Optimization(SimPO) demonstrating strong performance by eliminating the explicit reference model through a simple objective. However, the joint tuning of the hyperparameters $\beta$ and $\gamma$ in SimPO remains a central challenge. We argue that this difficulty arises because the margin formulation in SimPO is not easily interpretable across datasets with different reward gap structures. To better understand this issue, we conduct a comprehensive analysis of SimPO and find that $\beta$, rather than merely serving as a reward scaling factor, primarily functions to filter noisy samples, while the reward margin $\gamma$ defined by SimPO is not semantically well aligned with explicitly enlarging the reward gap between chosen and rejected responses. Motivated by these observations, we propose $\xi$-DPO: Direct preference optimization via ratio reward margin. We first reformulate the preference objective through an equivalent transformation, changing the optimization target from maximizing the likelihood of reward gaps to minimizing the distance between reward gaps and optimal margins. Then, we redefine the reward in a ratio form between the chosen and rejected, which effectively cancels the effect of $\beta$ and yields a bounded and interpretable margin. This margin is called the ratio reward margin and is denoted by $\xi$. Unlike the $\beta$ and $\gamma$ in SimPO, $\xi$ can be determined from the initial reward gap distribution, thus much easier to tune. Our extensive experimental results show that $\xi$-DPO substantially outperforms existing preference optimization methods across open benchmarks on multiple evaluation metrics.
Abstract (translated)
URL
https://arxiv.org/abs/2605.10981
PDF
-
HA-DiPO: Diffusion Policy Optimization for Heterogeneous Cooperative Transport
2026-10-24 10:24:37Xuebin Li, Jinpeng Xiao, Zong Chen, Pierluigi Arpenti, Bruno Siciliano, Jiebao Sun, Yiqun LiAbstract
Cooperative transport with heterogeneous legged robots requires coordinating different motion capabilities while supporting an unfixed shared payload. We present HA-DiPO (Heterogeneous-Agent Diffusion Policy Optimization), a hierarchical multi-agent reinforcement learning framework under centralized training with decentralized execution (CTDE). The framework decouples team-level coordination from embodiment-specific motion control: at the high level, a centralized critic uses team-value estimates to jointly train independent coordination policies, aligning robot-specific action responses with a shared transport objective; at the low level, embodiment-specific robust motion policies track the coordination commands. The high-level diffusion policies offer more flexible action distributions than single-Gaussian parameterizations, allowing potentially multimodal coordination responses to be represented. The coordination policies learn from scratch through reward-driven simulator interaction without demonstrations. At execution, each robot generates coordination commands using only local robot and payload observations, without partner-state inputs or inter-robot messages. Simulation comparisons show that HA-DiPO achieves higher final mean returns and velocity-tracking success rates, as well as lower velocity errors, than the evaluated baselines. Real-world experiments further demonstrate quadrupedal-wheeled-legged and human-robot cooperative transport.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33584
PDF
-
Eat Your Precision: Precision Objective Mitigates Reward Hacking in Rubric-based RL
2026-10-24 07:43:06Hyun Ryu, Jong-Heon Kim, Yiqing Xie, Seungone Kim, Carolyn Rose, Eunho YangAbstract
Rubric-based rewards enable effective reinforcement learning on non-verifiable tasks. However, because these rewards are typically defined as a weighted measure of rubric coverage, models are trained primarily to maximize a recall objective. As a result, RL-trained models produce responses that are more complete but also more verbose and contain more unnecessary or factually incorrect content, leading them to be less preferred than responses from the base model. To address this issue, we present the FORK reward, which jointly optimizes recall and a novel precision objective, which requires each unit of a model response to be both necessary and factually correct, as measured by rubric-driven necessity credits and a factual-correctness grader, respectively. Our experimental results show that FORK outperforms existing rubric-based RL methods and achieves Pareto-optimal performance that balances rubric coverage and overall response quality. Compared to the original recall-based reward, FORK substantially improves precision and reduces response length while achieving similar rubric coverage, which leads to significantly higher win rate compared to the base policy (e.g., from 10.6% to 68.0% on RubricHub-Medical and from 22.1% to 74.1% on HealthBench).
Abstract (translated)
URL
https://arxiv.org/abs/2610.33512
PDF
-
Learning Dexterous Manipulation: Hands, Simulators, Training, and Evaluation
2026-10-24 06:26:18Luai AbuelsamenAbstract
A dexterous hand can change an object's pose without setting it down. This survey covers 113 learning-based methods that attempt it, the hands they run on, the simulators they train in, and how they report results. The field has largely converged on one recipe: reinforcement learning on privileged state in a GPU simulator, distilled into a vision-only student for transfer to hardware. Almost half the corpus trains that way, nearly all of it in one simulator, whose own successors are barely used. A second and faster-growing branch learns from teleoperated demonstration instead, through vision-language-action models, diffusion policies, and systems whose contribution is the collection rig itself. Published work has tripled since 2022. Grasping remains the largest task family and bimanual coordination the fastest-rising, and almost every experiment runs on one of four hands. What the survey adds comes from reading released code and vendor specifications beside the papers. Eight hands that can be bought or built from published designs appear in no method paper here. Where the depth a hand sinks into the object it holds is reported at all, it is not measured in the physics that produced the motion. Section VII proposes a protocol for the unreported quantities; the body gives the counts behind every claim above.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33486
PDF
-
MDPO: Multi-constraint Dexterous Policy Optimization with Reinforcement Learning
2026-10-24 06:09:14Ao Zhou, Zongyou Yu, Ke Wu, Tao Xu, Junqiao Zhao, Wenchao Ding, Alois Knoll, Guang ChenAbstract
Dexterous manipulation requires high-dimensional control policies to accomplish complex contact-rich tasks while simultaneously satisfying multiple safety requirements, such as joint limits, unsafe finger configurations, arm collisions, and unstable object interactions. Constrained reinforcement learning (RL) provides a principled framework for incorporating these requirements into dexterous manipulation policy learning. However, existing constrained RL methods for dexterous manipulation primarily focus on cumulative cost constraints and struggle to simultaneously handle multiple heterogeneous constraints during dexterous policy learning. In this work, we propose Multi-constraint Dexterous Policy Optimization (MDPO) for safe dexterous manipulation. We first formulate dexterous manipulation as a multi-constraint policy optimization problem involving both cumulative cost constraints and state-wise cost constraints. Based on this formulation, we integrate multi-constraint projection into the dexterous manipulation policy optimization process. Specifically, we employ constraint-specific policy projections to map the intermediate policy that maximizes task performance onto the feasible regions defined by the cumulative cost constraint and the state-wise cost constraint. Extensive experiments on ReDMan, including safe joint control, finger safety, and grasping tasks, demonstrate that MDPO achieves superior constraint satisfaction while maintaining competitive task performance compared with the baselines. Our code is available at this link.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33480
PDF
-
Qwen-Image-PE: Learning Visual Design for Text-to-Image Prompt Enhancement
2026-10-24 05:50:40Zekai Zhang, Yanran Zhang, Shengming Yin, Yan Shu, Zhening Liu, Bingxuan Zhao, Chenyang Zhu, Donghao Zhou, Jiahao Li, Jiahao Wang, Jie Zhang, Kaiyuan Gao, Kun Yan, Lihan Jiang, Ningyuan Tang, Qihang Ai, Rui Song, Shuo Cao, Xiao Xu, Xiaoyue Chen, Yuxiang Chen, Zhendong Wang, Zhouyang Li, Zihao Liu, Zikai Zhou, Bo Zheng, Huishuai Zhang, Dongyan Zhao, Chenfei WuAbstract
Learning visual design is essential for prompt enhancement (PE) to turn underspecified user requests into coherent visual specifications for text-to-image (T2I) generation. Existing methods preserve user intent and improve overall visual appeal, with less explicit emphasis on fine-grained design dimensions. To address this gap, we introduce Qwen-Image-PE, which learns visual design through two-stage training. Supervised fine-tuning with Caption-Based Reverse Distillation (CRD) uses dense captions of well-designed images as design supervision. Reinforcement learning with Groupwise Design Reward (GDR) jointly compares all candidate images for the same request to refine lighting, composition, layout, and detail. On Qwen-Image-Bench with Qwen-Image-2.1, Qwen-Image-PE-397B improves over original requests by 7.81 points, while Qwen-Image-PE-9B outperforms all evaluated PE baselines. Further evaluations show gains in visual design, aesthetics, image quality, and human preference. We find that PE scales along three axes: training data, model size, and information density, with sufficient information density as a prerequisite for effective enhancement. Moreover, we demonstrate that Qwen-Image-PE generalizes to four additional T2I backbones, even surpassing native prompt enhancers on HunyuanImage-2.1 and FLUX 3 Image Klein. We release Qwen-Image-PE-9B at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33472
PDF
-
A Comprehensive Framework for All Stages of Large-Model Training: From Data and Tokenization, through Pretraining, to Post-Training Alignment
2026-10-24 05:09:05Xiangjing An, Zhibin Jiao, Aiting ChenAbstract
The capabilities of large language models (LLMs) and multimodal LLMs (MLLMs) arise from the synergy of the full training pipeline, including data and tokenization, pretraining, and post-training (supervised fine-tuning, preference alignment, reinforcement learning, and continual optimization), rather than any individual stage. Existing surveys mostly adopt a single-stage or single-technique perspective, treating related research as isolated systems without unified stage division or a horizontally comparable basis. This paper proposes a unified stage division and a six-dimensional analytical framework covering data, objective function, optimization strategy, system constraints, evaluation, and risk and governance. We systematically review 395 core references from 2017 to 2026, and present a conceptual map of cross-stage effects: forward constraints of pretraining on post-training, and reverse empowerment of post-training on data and pretraining. We derive three core findings. First, the comparability of scaling laws depends on data quality and mixture, a long-neglected variable. Second, preference alignment methods (RLHF, DPO) and reasoning reinforcement learning with verifiable rewards redefine post-training boundaries, while the Pass@1 versus Pass@K debate remains unresolved. Third, multimodal training poses structural challenges to the pure-text paradigm in tokenizer design, training objectives, and evaluation. The paper provides a cross-stage effects diagram, a method-family map, a practical decision tree, and at least 10 verifiable open problems, with an extensible skeleton for focused sub-surveys per stage. As of the retrieval date, peer-reviewed evidence (A+B) accounts for approximately 79% of the 395 references.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33451
PDF
-
Of Chains and Gains: What Reinforcement Learning changes on Reasoning-based Image Quality Assessment
2026-10-23 18:49:15Yixuan Li, Yuxuan Jiang, Hanwei Zhu, Hancheng Zhu, Fan Zhang, Shiqi Wang, David BullAbstract
Reinforcement Learning with Verifiable Rewards (RLVR) has improved the performance of large Vision-Language Models (VLMs) on image quality assessment (IQA). Existing approaches often attribute these gains to improved reasoning before scoring, supported by increasingly plausible Chain-of-Thought (CoT). However, whether these gains exceed sampling effects and how they depend on explicit reasoning remain unclear. We rigorously investigate these questions through sampling-aware evaluation, controlled reasoning interventions, and shortcut experiments. We find that RL-trained models outperform base models under large-budget $\operatorname{max}_g@k$ evaluation, while improvements beyond averaging are concentrated in-distribution. RLVR produces chains that raise base-model performance when transplanted and improves scoring with the chain held fixed. Yet scoring becomes less dependent on the chain during inference, more clearly at 7B than at 32B. Steering the probed quality direction in thinking tokens also has little effect on scoring. Finally, RL can improve scoring accuracy by exploiting textual cues correlated with mean opinion scores (MOSs), which could inflate benchmark performance. Together, these findings show that RLVR raises quality scoring and produces more useful chains, but that better scoring does not imply a stronger reliance on explicit reasoning, while score-based rewards can also encourage reliance on quality-irrelevant shortcuts. We hope this preliminary study can enlighten more and more concrete reasoning schemes in IQA.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33203
PDF
