Reward Modeling

Ask about this topic

Overview

Reward modeling turns comparisons, ratings, demonstrations, rubrics, or verifiable outcomes into signals for judging responses and actions. Methods range from pairwise preference models and scalar outcome or process critics to generative rubric judges, tool-based verifiers, and uncertainty-aware or personalized models. These signals guide preference optimization, best-of-N selection, test-time search, and training in language, image and video generation, robotics, and agents. A central challenge is that high preference accuracy may not capture diverse or shifting human intent, while optimization can exploit scoring biases, so evaluation must consider calibration and downstream behavior as well as ranking accuracy.

Latest papers

CardsList

325 papers49 papers in the last four weeks, up 69% on the four weeks before. 0.3% of all new papers.

May 3, 2026

RMGAP: Benchmarking the Generalization of Reward Models across Diverse Preferences

Reinforcement Learning from Human Feedback has become the standard paradigm for language model alignment, where reward models directly determine alignment effectiveness. In this work, we focus on how to evaluate the generalizability of reward models. By "generalizability", we mean the ability of RMs to correctly rank responses to align with diverse user preferences. However, existing reward model benchmarks are typically designed around a universal preference, failing to assess this generalization. To address this critical gap, we introduce RMGAP, a benchmark comprising 1,097 instances across Chat, Writing, Reasoning, and Safety domains. Since different users exhibit diverse preferences for the same task, we first generate four distinct responses with different linguistic profiles for each collected prompt. However, the original prompt set lacks the specificity to convey different preferences. We therefore construct tailored prompts by contrasting these candidates and designing scenarios in which one response becomes the uniquely appropriate choice. Moreover, we observe that users often express the same preference using different phrasings, and thus extend each prompt with two paraphrased variants. Our evaluation of 24 state-of-the-art RMs reveals their substantial limitations: even the best RM achieves only 49.27% Best-of-N accuracy, highlighting considerable room for improvement in reward model generalization. Related data and code are available at https://github.com/nanzhi84/RMGAP.

Yangyang Zhou, Yi-Chen Li

Beijing University of Posts and Telecommunications · National Key Laboratory for Novel Software Technology, Nanjing University, China · School of Artificial Intelligence, Nanjing University, China

Reward ModelingLLM EvaluationPreference ModelingReward Model Evaluation
AskPDF
May 2, 2026

Adaptive Pluralistic Alignment: A pipeline for dynamic artificial democracy

Prevailing alignment methods target a fixed set of preferences and therefore risk forcing value lock-in as societal norms evolve over time. We introduce Adaptive Pluralistic Alignment (APA), a modular pipeline for updating pluralistically aligned AI systems to track evolving values and avoid value lock-in without repeating costly pretraining or large-scale data collection. APA has three stages: (1) learning compact personalized reward models via low-rank reward basis decomposition, (2) using these models as a jury that collectively selects among candidate outputs through social-choice-theoretic voting, and (3) efficiently adapting the jury over time by fitting new annotator weights over the fixed reward bases as values shift. The resulting system is efficient, explainable, steerable, and modular. We implement a proof-of-concept instantiation using simulated historical annotators, demonstrate how jury composition and the choice of voting rule can substantially affect outcomes, and bound value tracking error given a sliding-window jury. We provide full code at https://github.com/RachelFreedman/apa.

Rachel Freedman

Department of Electrical Engineering and Computer Science, University of California, Berkeley, USA.

Reward ModelingLLM AlignmentPluralistic Alignment
AskPDF
May 1, 2026

Themis: Training Robust Multilingual Code Reward Models for Flexible Multi-Criteria Scoring

Reward models (RMs) have become an indispensable fixture of the language model (LM) post-training playbook, enabling policy alignment and test-time scaling. Research on the application of RMs in code generation, however, has been comparatively sparse, with existing work largely focusing on execution feedback. This choice constrains post-training to optimizing functional correctness over self-contained executable code. In this work, we examine the training and evaluation of multilingual, multi-criteria code RMs. To this end, we first compile Themis-CodeRewardBench, a benchmark to evaluate code RMs across five preference dimensions (i.e., criteria) and eight programming languages, on which we profile 50+ code, math, and general-purpose RMs. Observing the limited proficiency of current RMs beyond scoring for functional correctness, we develop Themis-CodePreference, the largest open-source collection of code preferences to date (more than 350k preference pairs), and use it to train Themis-RM, a suite of multilingual code reward models for flexible multi-criteria scoring, ranging in size from 600M to 32B parameters. Our experiments and ablations demonstrate positive scaling trends, strong cross-lingual transfer when training on diverse preferences, and the importance of multi-criteria training for reliable code reward modeling.

Indraneil Paul, Goran Glavaš, Iryna Gurevych

UKP Lab, TU Darmstadt and National Research Center for Applied Cybersecurity ATHENE · Center for Artificial Intelligence and Data Science, University of Würzburg

Reward ModelingReward Model Evaluation
AskPDF
May 1, 2026

PrefMoE: Robust Preference Modeling with Mixture-of-Experts Reward Learning

Preference-based reinforcement learning offers a scalable alternative to manual reward engineering by learning reward structures from comparative feedback. However, large-scale preference datasets, whether collected from crowdsourced annotators or generated by synthetic teachers, often contain heterogeneous and partially conflicting supervision, including disagreement across annotators and inconsistency within annotators. Existing reward learning methods typically fit a single reward model to such data, forcing it to average incompatible signals and thereby limiting robustness. To solve this, we propose PrefMoE, a mixture-of-experts reward learning framework for robust preference modeling. PrefMoE learns multiple specialized reward experts and uses trajectory-level soft routing to combine them adaptively, enabling the model to capture diverse latent preference patterns under noisy and heterogeneous preference supervision. A load-balancing regularizer further stabilizes training by preventing expert collapse. Across locomotion benchmarks from D4RL and manipulation tasks from MetaWorld, PrefMoE improves preference prediction robustness and leads to more reliable downstream policy learning than strong single-model baselines.

Ziqin Yuan, Ruiqi Wang, Dezhong Zhao +2

1Purdue University, West Lafayette, IN, USA. · 2Beijing University of Chemical Technology, Beijing, China. · 3Indiana University Bloomington, Bloomington, IN, USA.

Reward ModelingMixture of ExpertsPreference LearningPreference-Based RL
AskPDF
Apr 30, 2026

AesRM: Improving Video Aesthetics with Expert-Level Feedback

Despite rapid advances in photorealistic video generation, real-world applications such as filmmaking require video aesthetics, e.g., harmonious colors and cinematic lighting, beyond visual fidelity. Prior work on visual aesthetics largely focuses on images, often reducing aesthetics to coarse definitions, e.g., visual pleasure, without a rigorous and systematic evaluation. To improve video aesthetics, we propose a hierarchical rubric that decomposes video aesthetics into three core dimensions, Visual Aesthetics (VA), Visual Fidelity (VF), and Visual Plausibility (VP), with 15 fine-grained criteria, e.g., shot composition. This framework enables a large-scale expert-annotated preference dataset and an evaluation benchmark, AesVideo-Bench, containing about 2500 video pairs with expert annotations on VA, VF, and VP. We then build a family of Video Aesthetic Reward Models (AesRM): AesRM-Base, which directly predicts pairwise preferences on these dimensions to provide efficient post-training rewards, and AesRM-CoT, which additionally generates CoT aligned with all 15 criteria to improve assessment interpretability. Specifically, we train AesRM with a three-stage progressive scheme: (1) Atomic Aesthetic Capability Learning, which strengthens AesRM's recognition of fundamental aesthetic concepts, e.g., accurately identifying centered composition; (2) Cold-Start, aligning the model with structured reasoning protocols; and (3) GRPO, further improving evaluation accuracy. To enhance AesRM-CoT, we additionally propose self-consistency-based CoT synthesis to improve CoT quality and design CoT-based process rewards during GRPO. Extensive experiments show AesRM outperforms baselines on multiple aesthetics benchmarks and is more robust, with lower position bias. Finally, we align Wan2.2 with AesRM and observe clear aesthetic gains over existing aesthetic reward models.

Yujin Han, Yujie Wei, Yefei He +7

The University of Hong Kong · Alibaba Group · Fudan University +3

Reward ModelingVideo Quality AssessmentPreference LearningVideo Generation Evaluation
AskPDF
Apr 30, 2026

Leveraging Verifier-Based Reinforcement Learning in Image Editing

While Reinforcement Learning from Human Feedback (RLHF) has become a pivotal paradigm for text-to-image generation, its application to image editing remains largely unexplored. A key bottleneck is the lack of a robust general reward model for all editing tasks. Existing edit reward models usually give overall scores without detailed checks, ignoring different instruction requirements and causing biased rewards. To address this, we argue that the key is to move from a simple scorer to a reasoning verifier. We introduce Edit-R1, a framework that builds a chain-of-thought (CoT) verifier-based reasoning reward model (RRM) and then leverages it for downstream image editing. The Edit-RRM breaks instructions into distinct principles, evaluates the edited image against each principle, and aggregates these checks into an interpretable, fine-grained reward. To build such an RRM, we first apply supervised fine-tuning (SFT) as a ``cold-start'' to generate CoT reward trajectories. Then, we introduce Group Contrastive Preference Optimization (GCPO), a reinforcement learning algorithm that leverages human pairwise preference data to reinforce our pointwise RRM. After building the RRM, we use GRPO to train editing models with this non-differentiable yet powerful reward model. Extensive experiments demonstrate that our Edit-RRM surpasses powerful VLMs such as Seed-1.5-VL and Seed-1.6-VL as an editing-specific reward model, and we observe a clear scaling trend, with performance consistently improving from 3B to 7B parameters. Moreover, Edit-R1 delivers gains to editing models like FLUX.1-kontext, highlighting its effectiveness in enhancing image editing.

Hanzhong Guo, Jie Wu, Jie Liu +6

School of Computing and Data Science, The University of Hong Kong · ByteDance Seed · CUHK +1

Reward ModelingReinforcement LearningRL for Diffusion ModelsRL for Image Editing
AskPDF
Apr 30, 2026

Debiasing Reward Models via Causally Motivated Inference-Time Intervention

Reward models (RMs) play a central role in aligning large language models (LLMs) with human preferences. However, RMs are often sensitive to spurious features such as response length. Existing inference-time approaches for mitigating these biases typically focus exclusively on response length, resulting in performance trade-offs. In this paper, we propose causally motivated intervention for mitigating multiple types of biases in RMs at inference time. Our method first identifies neurons whose activations are strongly correlated with predefined bias attributes, and applies neuron-level intervention that suppresses these signals. We evaluate our method on RM benchmarks and observe reductions in sensitivity to spurious features across diverse bias types, without inducing performance trade-offs. Moreover, when used for preference annotation, small RMs (2B and 7B) with our method, which edits less than 2% of all the neurons in RMs, enable LLMs to improve alignment, achieving performance comparable to that of a state-of-the-art 70B RM on AlpacaEval and MT-Bench. Further analysis reveals that bias signals are primarily encoded by neurons in early layers, shedding light on the internal mechanisms of bias exploitation in RMs.

Kazutoshi Shinoda, Kosuke Nishida, Kyosuke Nishida

Human Informatics Labs., NTT, Inc.

Reward ModelingLLM AlignmentDebiased MLLanguage Model Bias EvaluationCausal Interventions in Language Models
AskPDF
Apr 30, 2026

From Coarse to Fine: Benchmarking and Reward Modeling for Writing-Centric Generation Tasks

Large language models have achieved remarkable progress in text generation but still struggle with generative writing tasks. In terms of evaluation, existing benchmarks evaluate writing reward models coarsely and fail to measure performance from the perspective of specific requirements. In terms of training, existing training methods either use LLM-as-a-judge approaches or train coarse-grained reward models, lacking fine-grained requirement-adherence reward modeling. To address these issues, we propose a fine-grained evaluation pipeline WEval for writing reward models and a fine-grained reinforcement learning training framework WRL. The evaluation data of WEval covers multiple task categories and requirement types, enabling systematic evaluation of writing reward models by measuring the correlation between the rankings of the reward model and gold rankings. WRL constructs positive and negative samples by selectively dropping instruction requirements, allowing for more precise reward model training. Experiments show that our models achieve substantial improvements across various writing benchmarks and exhibit strong generalization. The code and data are publicly available at https://github.com/Rainier-rq1/From_Coarse_to_Fine.

Qingyu Ren, Tianjun Pan, Xingzhou Chen +1

Shanghai Key Laboratory of Data Science, College of Computer Science and Artificial Intelligence, Fudan University · 2Shanghai Artificial Intelligence Laboratory

Reward ModelingRL for Language ModelsLLM EvaluationLanguage Model Generation EvaluationInstruction FollowingReward Model Evaluation
AskPDF
Apr 28, 2026

reward-lens: A Mechanistic Interpretability Library for Reward Models

Every RLHF-trained language model is shaped by a reward model, yet the mechanistic interpretability toolkit -- logit lens, direct logit attribution, activation patching, sparse autoencoders -- was built for generative LLMs whose primitives all project onto a vocabulary unembedding. Reward models replace that with a scalar regression head, breaking each tool. We present reward-lens, an open-source library that ports this toolkit to reward models, organised around one observation: the reward head's weight vector wrw_r is the natural axis for every interpretability question. The library provides a Reward Lens, component attribution, three-mode activation patching, a reward-hacking probe suite, TopK SAE feature attribution, cross-model comparison, and five theory-grounded extensions (distortion index, divergence-aware patching, misalignment cascade detection, reward-term conflict analysis, concept-vector analysis). A ten-method adapter protocol covers Llama, Mistral, Gemma-2, and ArmoRM multi-objective heads, with a generic adapter for any HuggingFace sequence classification model. We validate on two production reward models across ~695 RewardBench pairs. The central empirical finding is negative: linear attribution does not predict causal patching effects (mean Spearman ρ=−0.256ρ= -0.256 on Skywork, −0.027-0.027 on ArmoRM). The framework treats this disagreement as a property to expose, not a bug -- motivating a design that keeps observational and causal views first-class and directly comparable.

Mohammed Suhail B Nadaf

Reward ModelingLLM InterpretabilityCausal Interventions in Language Models
AskPDF
Apr 28, 2026

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient

Training language models via reinforcement learning often relies on imperfect proxy rewards, since ground truth rewards that precisely define the intended behavior are rarely available. Standard metrics for assessing the quality of proxy rewards, such as ranking accuracy, treat incorrect rewards as strictly harmful. In this work, however, we highlight that not all deviations from the ground truth are equal. By theoretically analyzing which outputs attract probability during policy gradient optimization, we categorize reward errors according to their effect on the increase in ground truth reward. The analysis establishes that reward errors, though conventionally viewed as harmful, can also be benign or even beneficial by preventing the policy from stalling around outputs with mediocre ground truth reward. We then present two practical implications of our theory. First, for reinforcement learning from human feedback (RLHF), we develop reward model evaluation metrics that account for the harmfulness of reward errors. Compared to standard ranking accuracy, these metrics typically correlate better with the performance of a language model after RLHF, yet gaps remain in robustly evaluating reward models. Second, we provide insights for reward design in settings with verifiable rewards. A key theme underlying our results is that the effectiveness of a proxy reward function depends heavily on its interaction with the initial policy and learning algorithm.

Shuning Shang, Hubert Strauss, Stanley Wei +2

Princeton Language and Intelligence, Princeton University

Reward ModelingPolicy GradientPolicy LearningReward Design for RLReward Model EvaluationRL from Human Feedback
AskPDF
Apr 27, 2026

Model-Free Inference of Investor Preferences: A Relative Entropy IRL Approach

We present a framework using Relative Entropy Inverse Reinforcement Learning (RE-IRL) to recover investor reward functions from observed investment actions and market conditions. Unlike traditional IRL algorithms, RE-IRL is employed to account for environments where transition probabilities are unknown or inaccessible. To address the challenge of data sparsity, we utilize a KK-nearest neighbor approach to estimate the observed behavior policy. Furthermore, we propose a statistical testing framework to evaluate the validity and robustness of the estimated results.

Chen Xu

Department of Engineering, Shenzhen MSU-BIT University. Address: 1 International University Park Road, Longgang District, Shenzhen, Guangdong, China.

Reward ModelingReinforcement LearningFinancial Decision-Making
AskPDF
Apr 27, 2026

Explanation Quality Assessment as Ranking with Listwise Rewards

We reformulate explanation quality assessment as a ranking problem rather than a generation problem. Instead of optimizing models to produce a single "best" explanation token-by-token, we train reward models to discriminate among multiple candidate explanations and learn their relative quality. Concretely, we construct per-instance candidate sets with graded quality levels and train listwise and pairwise ranking models (ListNet, LambdaRank, RankNet) to preserve ordinal structure and avoid score compression typical of pointwise regression or binary preference objectives. We observe three findings: First, ranking losses consistently outperform regression on score separation across all domains tested. Second, the optimal ranking loss depends on data characteristics: listwise objectives excel with well-separated quality tiers, while pairwise methods are more robust to noisy natural annotations. Third, when trained on carefully curated and well-structured data, small encoder models can match models that are orders of magnitude larger, suggesting that data quality matters more than model scale. Finally, when used as rewards in policy optimization, ranking-based scores enable stable convergence in settings where regression-based rewards fail entirely. Code and data are available at: https://github.com/Tankiit/PPO_Learning_to_rank

Thomas Bailleux, Tanmoy Mukherjee, Emmanuel Lonca +2

CRIL, Univ. Artois & CNRS, France · Institut Universitaire de France

Reward ModelingLearning to RankPairwise RankingExplainability Evaluation
AskPDF
Apr 24, 2026

Reward Models Are Secretly Value Functions: Temporally Coherent Reward Modeling

Reward models in RLHF are trained to score only the final token of a response - a choice that discards rich signal from every intermediate position and produces models whose token-level outputs are noise. We argue this is a missed opportunity: a well-trained reward model's output at any token should represent the conditional expectation of the final reward given the response so far. We introduce Temporally Coherent Reward Modeling (TCRM), which induces this property via two regularization terms on top of the standard Bradley-Terry loss, with minimizers provably equal to conditional expectations. The regularizers correspond to Monte Carlo and TD value-learning objectives, establishing a direct connection to RL value functions. TCRM requires zero changes to architecture, data, or inference, yet unlocks three capabilities from one principle: interpretable token-level reward trajectories (middle-token pairwise accuracy improved from 50% to 88.9%, final-token accuracy preserved); state-of-the-art PRM performance on ProcessBench (44.9% average F1) among models trained only on outcome data; and unified reward/value modeling in PPO, reducing peak GPU memory by 27% and step time by 19% with matching LLM quality.

Alex Nikulkov

AI at Meta

Reward ModelingValue Function EstimationProcess Reward ModelsRL Post-Training
AskPDF
Apr 20, 2026

ARES: Adaptive Red-Teaming and End-to-End Repair of Policy-Reward System

Reinforcement Learning from Human Feedback (RLHF) is central to aligning Large Language Models (LLMs), yet it introduces a critical vulnerability: an imperfect Reward Model (RM) can become a single point of failure when it fails to penalize unsafe behaviors. While existing red-teaming approaches primarily target policy-level weaknesses, they overlook what we term systemic weaknesses cases where both the core LLM and the RM fail in tandem. We present ARES, a framework that systematically discovers and mitigates such dual vulnerabilities. ARES employs a ``Safety Mentor'' that dynamically composes semantically coherent adversarial prompts by combining structured component types (topics, personas, tactics, goals) and generates corresponding malicious and safe responses. This dual-targeting approach exposes weaknesses in both the core LLM and the RM simultaneously. Using the vulnerabilities gained, ARES implements a two-stage repair process: first fine-tuning the RM to better detect harmful content, then leveraging the improved RM to optimize the core model. Experiments across multiple adversarial safety benchmarks demonstrate that ARES substantially enhances safety robustness while preserving model capabilities, establishing a new paradigm for comprehensive RLHF safety alignment.

Jiacheng Liang, Yao Ma, Tharindu Kumarage +5

Stony Brook University · Amazon Nova Responsible AI

Reward ModelingAdversarial Attacks on LLMsLLM Red TeamingLLM Safety AlignmentRL from Human Feedback
AskPDF
Apr 20, 2026

PARM: Pipeline-Adapted Reward Model

Reward models (RMs) are central to aligning large language models (LLMs) with human preferences, powering RLHF and advanced decoding strategies. While most prior work focuses on single-step generation, real-world applications increasingly adopt multi-stage LLM pipelines, where effective reward guidance remains underexplored. We investigate this through code generation for combinatorial optimization, constructing a pipeline that integrates reward models into both formulation and solution stages. We identify a critical challenge: inconsistency between reward model predictions and actual pipeline execution outcomes. To address this, we propose the Pipeline-Adapted Reward Model (PARM), which leverages pipeline-specific data and direct preference optimization to align rewards with downstream feedback. We instantiate PARM as a two-stage pipeline (formulation -> code generation) and evaluate it on four public optimization benchmarks, measuring execution rate and solving accuracy against baselines and sampling methods. A supplementary cross-domain experiment on GSM8K assesses transferability. Results demonstrate that PARM consistently improves pipeline output quality and stability, providing new insights into reward modeling for multi-stage LLM reasoning.

Xingyu Fan, Wei Shao, Jiacheng Liu +2

Reward ModelingLLM-Based Program Synthesis
AskPDF
Apr 20, 2026

QuantumQA: Enhancing Scientific Reasoning via Physics-Consistent Dataset and Verification-Aware Reinforcement Learning

Large language models (LLMs) show strong capabilities in general reasoning but typically lack reliability in scientific domains like quantum mechanics, which demand strict adherence to physical constraints. This limitation arises from the scarcity of verifiable training resources and the inadequacy of coarse feedback signals in standard alignment paradigms. To address the data challenge, we introduce QuantumQA, a large-scale dataset constructed via a task-adaptive strategy and a hybrid verification protocol that combines deterministic solvers with semantic auditing to guarantee scientific rigor. Building on this foundation, we propose the verification-aware reward model (VRM) tailored for Reinforcement Learning with Verifiable Rewards (RLVR), which employs an adaptive reward fusion (ARF) mechanism to dynamically integrate deterministic signals from a scientific execution suite (SES) with multidimensional semantic evaluations for precise supervision. Experimental results demonstrate that our method consistently outperforms baselines and general-purpose preference models. Notably, our optimized 8B model achieves performance competitive with proprietary models, validating that incorporating verifiable, rule-based feedback into the reinforcement learning loop offers a parameter-efficient alternative to pure scaling.

Songxin Qu, Tai-Ping Sun, Yun-Jie Wang +8

Institute of Advanced Technology, University of Science and Technology of China · Institute of Artificial Intelligence, Hefei Comprehensive National Science Center · School of Physics, University of Science and Technology of China +2

Reward ModelingScientific QAReinforcement Learning with Verifiable Rewards
AskPDF
Apr 19, 2026

TinyJudge: Unverifiable Constraint Alignment via Lightweight Specialist Ensembles

Instruction Following (IF) is a core capability of LLMs, requiring strict adherence to diverse constraints, ranging from verifiable ones (e.g., output length) to unverifiable ones (e.g., tone). Reinforcement learning with verifiable rewards has emerged as a paradigm for IF tasks, leveraging LLM-as-a-judge to assess unverifiable constraints. However, we empirically find that this approach remains a significant bottleneck, suffering from severe reward hacking and higher computational overhead. In this work, we first analyze the generalization capabilities of unverifiable constraints and discover that specific constraints exhibit distinct, high-generalization patterns. Motivated by this, we propose TinyJudge, a framework that employs an ensemble of specialized tiny language models (∼0.6B\sim0.6B) to provide rewards for soft constraints. By distilling expertise from frontier models into these tiny models, it achieves high-precision, lightweight evaluation. Extensive evaluations across five benchmarks demonstrate that TinyJudge outperforms the baselines by ∼10%\sim10\% in average performance and 12%12\% in reward precision. Crucially, it also achieves a 3×3\times speedup in total training time. Our work provides a scalable and robust path for aligning LLMs with unverifiable human instructions.

Yirong Zeng, Yufei Liu, Xiao Ding +9

Harbin Institute of Technology SCIR Lab · Peking University · Huawei Technologies Co., Ltd

Reward ModelingLLM AlignmentLLM-as-a-JudgeInstruction FollowingReinforcement Learning with Verifiable Rewards
AskPDF
Apr 17, 2026

AgentV-RL: Scaling Reward Modeling with Agentic Verifier

Verifiers have been demonstrated to enhance LLM reasoning via test-time scaling (TTS). Yet, they face significant challenges in complex domains. Error propagation from incorrect intermediate reasoning can lead to false positives for seemingly plausible solutions, while lacking external grounding makes verifiers unreliable on computation or knowledge-intensive tasks. To address these challenges, we propose Agentic Verifier, a framework that transforms reward modeling into a multi-turn, tool-augmented deliberative process. We introduce complementary forward and backward agents: one traces solutions from premises to conclusions, while the other re-checks conclusions against their underlying premises. This bidirectional process enables a comprehensive, reliable, and interpretable assessment of solutions. To facilitate practical deployment, we propose AgentV-RL. Through proactive exploration and reinforcement learning, the verifier autonomously interleaves tool-use with internal reasoning. Extensive experiments show that Agentic Verifier yields consistent performance gains under both parallel and sequential TTS. Notably, our 4B variant surpasses state-of-the-art ORMs by 25.2%, positioning it as a promising paradigm for agentic reward modeling.

Jiazheng Zhang, Ziche Fu, Zhiheng Xi +13

College of Computer Science and Artificial Intelligence, Fudan University · Huazhong University of Science and Technology · The University of Hong Kong +3

Reward ModelingRL for Language Model ReasoningLLM AgentsLLM Agent Training
AskPDF
Apr 15, 2026

Robust Reward Modeling for Large Language Models via Causal Decomposition

Reward models are central to aligning large language models, yet they often overfit to spurious cues such as response length and overly agreeable tone. Most prior work weakens these cues directly by penalizing or controlling specific artifacts, but it does not explicitly encourage the model to ground preferences in the prompt's intent. We learn a decoder that maps a candidate answer to the latent intent embedding of the input. The reconstruction error is used as a signal to regularize the reward model training. We provide theoretical evidence that this signal emphasizes prompt-dependent information while suppressing prompt-independent shortcuts. Across math, helpfulness, and safety benchmarks, the decoder selects shorter and less sycophantic candidates with 0.877 accuracy. Incorporating this signal into RM training in Gemma-2-2B-it and Gemma-2-9B-it increases RewardBench accuracy from 0.832 to 0.868. For Best-of-N selection, our framework increases length-controlled win rates while producing shorter outputs, and remains robust to lengthening and mild off-topic drift in controlled rewrite tests.

Yunsheng Lu, Zijiang Yang, Licheng Pan +1

College of Computer Science and Technology, Zhejiang University · Department of Statistics, University of Chicago

Reward ModelingLLM AlignmentLanguage Model Robustness
AskPDF
Apr 15, 2026

C2: Scalable Rubric-Augmented Reward Modeling from Binary Preferences

Rubric-augmented verification guides reward models with explicit evaluation criteria, yielding more reliable judgments than single-model verification. However, most existing methods require costly rubric annotations, limiting scalability. Moreover, we find that rubric generation is vulnerable to a failure of cooperation; low-quality rubrics actively mislead reward models rather than help. Inspired by the principle of cooperative communication, we propose Cooperative yet Critical reward modeling (C2), a framework that significantly improves reward model judgments by having the reward model critically collaborate with a rubric generator trained solely from binary preferences. In C2, we synthesize helpful and misleading rubric pairs by measuring how each rubric shifts the reward model toward or away from the correct preference. Using these contrastive pairs, we train a cooperative rubric generator to propose helpful rubrics, and a critical verifier to assess rubric validity before making its judgment, following only rubrics it deems helpful at inference time. C2 outperforms reasoning reward models trained on the same binary preferences, with gains of up to 6.5 points on RM-Bench and 6.0 points length-controlled win rate on AlpacaEval 2.0. Without external rubric annotations, C2 enables an 8B reward model to match performance achieved with rubrics from a 4×\times larger model. Overall, our work demonstrates that eliciting deliberate cooperation in rubric-augmented verification makes reward models more trustworthy in a scalable way.

Akira Kawabata, Saku Sugawara

The Graduate University for Advanced Studies (SOKENDAI) · National Institute of Informatics · The Asahi Shimbun Company +1

Reward ModelingRubric-Based Evaluation
AskPDF
Apr 15, 2026

From Prediction to Justification: Aligning Sentiment Reasoning with Human Rationale via Reinforcement Learning

While Aspect-based Sentiment Analysis (ABSA) systems have achieved high accuracy in identifying sentiment polarities, they often operate as "black boxes," lacking the explicit reasoning capabilities characteristic of human affective cognition. Humans do not merely categorize sentiment; they construct causal explanations for their judgments. To bridge this gap, we propose ABSA-R1, a large language model framework designed to mimic this ``reason-before-predict" cognitive process. By leveraging reinforcement learning (RL), ABSA-R1 learns to articulate the why behind the what, generating natural language justifications that ground its sentiment predictions. We introduce a Cognition-Aligned Reward Model (formerly sentiment-aware reward model) that enforces consistency between the generated reasoning path and the final emotional label. Furthermore, inspired by metacognitive monitoring, we implement a performance-driven rejection sampling strategy that selectively targets hard cases where the model's internal reasoning is uncertain or inconsistent. Experimental results on four benchmarks demonstrate that equipping models with this explicit reasoning capability not only enhances interpretability but also yields superior performance in sentiment classification and triplet extraction compared to non-reasoning baselines.

Shihao Zhang, Ziwei Wang, Jie Zhou +6

School of Computer Science and Technology, East China Normal University · Shanghai Qiji Zhifeng Co., Ltd. · Ocean University of China

Aspect-Based Sentiment AnalysisRL for Language Model ReasoningReward Modeling
AskPDF
Apr 14, 2026

Bias at the End of the Score

Reward models (RMs) are inherently non-neutral value functions designed and trained to encode specific objectives, such as human preferences or text-image alignment. RMs have become crucial components of text-to-image (T2I) generation systems where they are used at various stages for dataset filtering, as evaluation metrics, as a supervisory signal during optimization of parameters, and for post-generation safety and quality filtering of T2I outputs. While specific problems with the integration of RMs into the T2I pipeline have been studied (e.g. reward hacking or mode collapse), their robustness and fairness as scoring functions remains largely unknown. We conduct a large scale audit of RM robustness with respect to demographic biases during T2I model training and generation. We provide quantitative and qualitative evidence that while originally developed as quality measures, RMs encode demographic biases, which cause reward-guided optimization to disproportionately sexualize female image subjects reinforce gender/racial stereotypes, and collapse demographic diversity. These findings highlight shortcomings in current reward models, challenge their reliability as quality metrics, and underscore the need for improved data collection and training procedures to enable more robust scoring.

Salma Abdel Magid, Grace Guo, Esin Tureci +4

Princeton University · Harvard University

Reward ModelingT2I GenerationDemographic Bias in T2I ModelsT2I Generation Evaluation
AskPDF
Apr 14, 2026

PromptEcho: Annotation-Free Reward from Vision-Language Models for Text-to-Image Reinforcement Learning

Reinforcement learning (RL) can improve the prompt following capability of text-to-image (T2I) models, yet obtaining high-quality reward signals remains challenging: CLIP Score is too coarse-grained, while VLM-based reward models (e.g., RewardDance) require costly human-annotated preference data and additional fine-tuning. We propose PromptEcho, a reward construction method that requires no annotation and no reward model training. Given a generated image and a guiding query, PromptEcho computes the token-level cross-entropy loss of a frozen VLM with the original prompt as the label, directly extracting the image-text alignment knowledge encoded during VLM pretraining. The reward is deterministic, computationally efficient, and improves automatically as stronger open-source VLMs become available. For evaluation, we develop DenseAlignBench, a benchmark of concept-rich dense captions for rigorously testing prompt following capability. Experimental results on two state-of-the-art T2I models (Z-Image and QwenImage-2512) demonstrate that PromptEcho achieves substantial improvements on DenseAlignBench (+26.8pp / +16.2pp net win rate), along with consistent gains on GenEval, DPG-Bench, and TIIFBench without any task-specific training. Ablation studies confirm that PromptEcho comprehensively outperforms inference-based scoring with the same VLM, and that reward quality scales with VLM size. We will open-source the trained models and the DenseAlignBench.

Jinlong Liu, Wanggui He, Peng Zhang +3

Alibaba Group · Zhejiang University

Reward ModelingRL for Image GenerationT2I Generation Evaluation
AskPDF
Apr 13, 2026

RationalRewards: Reasoning Rewards Scale Visual Generation Both Training and Test Time

Most reward models for visual generation reduce rich human judgments to a single unexplained score, discarding the reasoning that underlies preference. We show that teaching reward models to produce explicit, multi-dimensional critiques before scoring transforms them from passive evaluators into active optimization tools, improving generators in two complementary ways: at training time, structured rationales provide interpretable, fine-grained rewards for reinforcement learning; at test time, a Generate-Critique-Refine loop turns critiques into targeted prompt revisions that improve outputs without any parameter updates. To train such a reward model without costly rationale annotations, we introduce Preference-Anchored Rationalization (PARROT), a principled framework that recovers high-quality rationales from readily available preference data through anchored generation, consistency filtering, and distillation. The resulting model, RationalRewards (8B), achieves state-of-the-art preference prediction among open-source reward models, competitive with Gemini-2.5-Pro, while using 10-20x less training data than comparable baselines. As an RL reward, it consistently improves text-to-image and image-editing generators beyond scalar alternatives. Most strikingly, its test-time critique-and-refine loop matches or exceeds RL-based fine-tuning on several benchmarks, suggesting that structured reasoning can unlock latent capabilities in existing generators that suboptimal prompts fail to elicit.

Haozhe Wang, Cong Wei, Weiming Ren +3

HKUST · University of Waterloo · Alibaba

Reward ModelingMultimodal Reward ModelingRL for Image GenerationPrompt Optimization for T2I Generation
AskPDF
Apr 13, 2026

DDO-RM: Distribution-Level Policy Improvement after Reward Learning

Recent theory suggests that reward-model-first methods can be more sample-efficient than direct policy fitting when the reward function is statistically simpler than the induced policy. We propose DDO-RM, a finite-candidate decision-optimization method that converts reward scores into an explicit target distribution. Unlike PPO-based RLHF or DPO, DDO-RM performs a KL-regularized mirror-descent update to project the policy toward a reward-improved distribution over a candidate set. Preliminary experiments on Pythia-410M show that DDO-RM outperforms DPO in pair accuracy (0.52 to 0.56) and mean margin (0.13 to 0.53). Our framework provides a principled connection between reward learning and mirror-descent policy improvement.

Tiantian Zhang, Jierui Zuo, Michael Chen +1

Department of Computer Science Columbia University, New York, NY 10027, USA · Department of Management Science and Engineering Tsinghua University, Beijing, China · Department of Computer Science Carnegie Mellon University, Pittsburgh, PA 15213, USA

Policy OptimizationKL-Regularized RLReward Modeling
AskPDF
Apr 13, 2026

You Only Judge Once: Multi-response Reward Modeling in a Single Forward Pass

We present a discriminative multimodal reward model that scores all candidate responses in a single forward pass. Conventional discriminative reward models evaluate each response independently, requiring multiple forward passes, one for each potential response. Our approach concatenates multiple responses with separator tokens and applies cross-entropy over their scalar scores, enabling direct comparative reasoning and efficient NN-way preference learning. The multi-response design also yields up to N×N\times wall-clock speedup and FLOPs reduction over conventional single-response scoring. To enable NN-way reward evaluation beyond existing pairwise benchmarks, we construct two new benchmarks: (1) MR2^2Bench-Image contains human-annotated rankings over responses from 8 diverse models; (2) MR2^2Bench-Video is a large-scale video-based reward benchmark derived from 94K crowdsourced pairwise human judgments over video question-answering spanning 19 models, denoised via preference graph ensemble. Both benchmarks provide 4-response evaluation variants sampled from the full rankings. Built on a 4B vision-language backbone with LoRA fine-tuning and a lightweight MLP value head, our model achieves state-of-the-art results on six multimodal reward benchmarks, including MR2^2Bench-Image, MR2^2Bench-Video, and four other existing benchmarks. Our model outperforms existing larger generative and discriminative reward models. We further demonstrate that our reward model, when used in reinforcement learning with GRPO, produces improved policy models that maintain performance across standard multimodal benchmarks while substantially improving open-ended generation quality, outperforming a single-response discriminative reward model (RM) baseline by a large margin in both training stability and open-ended generation quality.

Yinuo Yang, Zixian Ma, Manasi Ganti +2

Paul G. Allen School of Computer Science & Engineering University of Washington Seattle, WA 98195, USA

Reward ModelingMultimodal Reward ModelingReward Model Evaluation
AskPDF
Apr 11, 2026

Reason Only When Needed: Efficient Generative Reward Modeling via Model-Internal Uncertainty

Recent advancements in the Generative Reward Model (GRM) have demonstrated its potential to enhance the reasoning abilities of LLMs through Chain-of-Thought (CoT) prompting. Despite these gains, existing implementations of GRM suffer from two critical limitations. First, CoT prompting is applied indiscriminately to all inputs regardless of their inherent complexity. This introduces unnecessary computational costs for tasks amenable to fast, direct inference. Second, existing approaches primarily rely on voting-based mechanisms to evaluate CoT outputs, which often lack granularity and precision in assessing reasoning quality. In this paper, we propose E-GRM, an efficient generative reward modeling framework grounded in model-internal uncertainty. E-GRM leverages the convergence behavior of parallel model generations to estimate uncertainty and selectively trigger CoT reasoning only when needed, without relying on handcrafted features or task-dependent signals. To improve reward fidelity, we introduce a lightweight discriminative scorer trained with a hybrid regression--ranking objective to provide fine-grained evaluation of reasoning paths. Experiments on multiple reasoning benchmarks show that E-GRM substantially reduces inference cost while consistently improving answer accuracy, demonstrating that model-internal uncertainty is an effective and general signal for efficient reasoning-aware reward modeling.

Chao Xue, Yao Wang, Mengqiao Liu +11

University of New South Wales, Australia · Tencent Hunyuan, China · Tencent Yuanbao, China +2

Reward ModelingEfficient Language Model ReasoningCoT Reasoning
AskPDF
Apr 10, 2026

Relational Preference Encoding in Looped Transformer Internal States

We investigate how looped transformers encode human preference, training lightweight evaluator heads on frozen Ouro-2.6B loop-iteration states on Anthropic HH-RLHF. v2: an erratum is prepended; the original manuscript is unchanged. A post-publication audit found the three headline results inflated by two independent evaluation errors. The 95.2% pairwise evaluator accuracy is a canonical-ordering artifact: the data were correctly split, but the evaluator learned to prefer the first-presented argument; its strict antisymmetrized accuracy on the full 8,552-pair test set is 63.9%. The 84.5% pairwise probe and the below-chance 21.75% pointwise probe were source-item leaks (orientation rows and pair partners crossing the train/test split); corrected pair-disjoint values are 56.5% and 54.2% -- above chance, so the "inverted polarity" finding is withdrawn. The central finding survives at much smaller magnitude: preference is decoded more accurately relationally than pointwise (paired +2.3 points, 95% CI [+1.3, +3.3]), and the antisymmetrized evaluator still beats the linear probe, but no corrected readout rivals end-to-end reward models. The methodological findings stand (constant-output degeneracy, flip test, swap-protocol metric deflation), with one correction: antisymmetrized accuracy, not antisymmetry correlation, certifies relational discrimination. The two errors are mutually invisible -- a split audit cannot see an ordering prior, antisymmetrization cannot see a leak -- so both checks are required. Full audit in the follow-up work (Kirin, 2026, in preparation).

Jan Kirin

Pairwise Preference LearningReward ModelingTransformer InterpretabilityPairwise ComparisonPairwise Preference EvaluationPairwise Ranking+2
AskPDF
Apr 9, 2026

ProMedical: Hierarchical Fine-Grained Criteria Modeling for Medical LLM Alignment via Explicit Injection

Aligning Large Language Models (LLMs) with high-stakes medical standards remains a significant challenge, primarily due to the dissonance between coarse-grained preference signals and the complex, multi-dimensional nature of clinical protocols. To bridge this gap, we introduce ProMedical, a unified alignment framework grounded in fine-grained clinical criteria. We first construct ProMedical-Preference-50k, a dataset generated via a human-in-the-loop pipeline that augments medical instructions with rigorous, physician-derived rubrics. Leveraging this corpus, we propose the Explicit Criteria Injection paradigm to train a multi-dimensional reward model. Unlike traditional scalar reward models, our approach explicitly disentangles safety constraints from general proficiency, enabling precise guidance during reinforcement learning. To rigorously validate this framework, we establish ProMedical-Bench, a held-out evaluation suite anchored by double-blind expert adjudication. Empirical evaluations demonstrate that optimizing the Qwen3-8B base model via ProMedical-RM-guided GRPO yields substantial gains, improving overall accuracy by 22.3% and safety compliance by 21.7%, effectively rivaling proprietary frontier models. Furthermore, the aligned policy generalizes robustly to external benchmarks, demonstrating performance comparable to state-of-the-art models on UltraMedical. We publicly release our datasets, reward models, and benchmarks to facilitate reproducible research in safety-aware medical alignment.

He Geng, Yangmin Huang, Lixian Lai +5

Xunfei Healthcare Technology Co., Ltd.

LLM AlignmentMedical LLMsMulti-Objective Language Model AlignmentReward ModelingClinical Language Model Evaluation
AskPDF
Apr 8, 2026

ReflectRM: Boosting Generative Reward Models via Self-Reflection within a Unified Judgment Framework

Reward Models (RMs) are critical components in the Reinforcement Learning from Human Feedback (RLHF) pipeline, directly determining the alignment quality of Large Language Models (LLMs). Recently, Generative Reward Models (GRMs) have emerged as a superior paradigm, offering higher interpretability and stronger generalization than traditional scalar RMs. However, existing methods for GRMs focus primarily on outcome-level supervision, neglecting analytical process quality, which constrains their potential. To address this, we propose ReflectRM, a novel GRM that leverages self-reflection to assess analytical quality and enhance preference modeling. ReflectRM is trained under a unified generative framework for joint modeling of response preference and analysis preference. During inference, we use its self-reflection capability to identify the most reliable analysis, from which the final preference prediction is derived. Experiments across four benchmarks show that ReflectRM consistently improves performance, achieving an average accuracy gain of +3.7 on Qwen3-4B. Further experiments confirm that response preference and analysis preference are mutually reinforcing. Notably, ReflectRM substantially mitigates positional bias, yielding +10.2 improvement compared with leading GRMs and establishing itself as a more stable evaluator. Our code is available at https://github.com/yuliangCarmelo/ReflectRM.

Kai Qin, Liangxin Liu, Yu Liang +7

Shenzhen International Graduate School, Tsinghua University · Baidu Inc., Beijing, China

Preference ModelingSelf-Reflection in Language Model AgentsReward ModelingLLM AlignmentBias in LLM-as-a-Judge
AskPDF