Datasets:

fineset-io
/

reward-modeling-papers

id string	sources list	title string	abstract string	authors list	categories list	fields_of_study list	published_date timestamp[s]	url string	pdf_url string	arxiv_id string	doi string	citation_count int64	influential_citation_count int64	has_code bool	code_url string	venue string	quality_score float64
1d23ad55a293d9710bb07530b83eb18ff03de7fbb9e034f0fd160323bcaca32c	[ "arxiv" ]	Uncertainty-Aware Reward Modeling for Stable RLHF	Reinforcement learning from human feedback (RLHF) aligns large language models by training reward models on preference data and optimizing policies to maximize predicted rewards. However, this pipeline faces two fundamental challenges: (1) reward models cannot signal when their predictions are unreliable, since they us...	[ "Licheng Pan", "Haocheng Yang", "Haoxuan Li", "Yichen Sun", "Yunsheng Lu", "Shijian Wang", "Lei Shen", "Yuan Lu", "Zhixuan Chu", "Hao Wang" ]	[ "cs.LG", "cs.AI" ]	[]	2026-06-18T00:00:00	https://arxiv.org/abs/2606.19818	https://arxiv.org/pdf/2606.19818v1	2606.19818	null	0	0	false	null	null	0.35
536f9f096daad2029c71a41bf0a3d9f383d01be39eeb4abd0fd9f53b086c5a6f	[ "arxiv" ]	Pareto Q-Learning with Reward Machines	We present Pareto Q-Learning with Reward Machines (PQLRM), a multi-objective reinforcement learning algorithm for tasks whose reward structure is specified by a set of reward machines (RMs). PQLRM combines Pareto Q-Learning (PQL), which maintains sets of vector-valued Q-estimates to approximate the Pareto front, with e...	[ "Arnaud Lequen", "Clément Legrand-Lixon", "Léo Saulières" ]	[ "cs.LG", "cs.AI" ]	[]	2026-06-17T00:00:00	https://arxiv.org/abs/2606.19134	https://arxiv.org/pdf/2606.19134v1	2606.19134	null	0	0	false	null	null	0.35
15bf71c08d26be78e94a18979b27c82d8cebf65daa6462ac70d4d81641d370cf	[ "arxiv" ]	Steerable Cultural Preference Optimization of Reward Models	It is essential for large language model (LLM) technology to serve many different cultural sub-communities in a manner that is acceptable to each community. However, research on LLM alignment has so far predominantly focused on predicting a unified response preference of annotators from certain regions. This paper aims...	[ "Minsik Oh", "Advit Deepak", "Sophie Wu", "Douwe Kiela", "Ekaterina Shutova" ]	[ "cs.CL", "cs.AI" ]	[]	2026-06-17T00:00:00	https://arxiv.org/abs/2606.18606	https://arxiv.org/pdf/2606.18606v1	2606.18606	null	0	0	true	https://github.com/minsik-ai/Steerable-Cultural-Preference	null	0.65
e53b309a3001159f948afa6b4c085d2dbad13379d42615da4f2af1c957f0bdca	[ "arxiv", "semantic_scholar" ]	A Unifying Lens on Reward Uncertainty in RLHF	Reinforcement learning from human feedback (RLHF) is bottlenecked by reward hacking, where the policy exploits errors in a proxy reward model (RM) and produces high RM scores without genuine quality gains. A natural mitigation is pessimism: lowering rewards in regions where the RM is uncertain. However, standard scalar...	[ "Ely Hahami", "Yoel Zimmermann", "Ray Zhou", "Jack Benarroch Jedlicki" ]	[ "cs.LG", "cs.AI", "cs.CL" ]	[ "Computer Science" ]	2026-06-08T00:00:00	https://arxiv.org/abs/2606.09073	https://arxiv.org/pdf/2606.09073v2	2606.09073	null	0	0	false	null	null	0.35
2901871b1fafcfd85fc43399f35d27995799b32c96096606d7777709fe9cb5c5	[ "arxiv", "semantic_scholar" ]	DynaCF: Mitigating Shortcut Learning in Reward Models via Dynamic Counterfactual Sensitivity	Reward models trained from pairwise preferences often exploit superficial shortcut cues rather than learning true response quality. We propose DynaCF, a dynamic reweighting framework for mitigating shortcut learning in reward model training. Unlike static shortcut heuristics, DynaCF measures shortcut sensitivity online...	[ "Fengyuan Liu", "Yongliang Miao", "Zirui He", "Yanguang Liu", "Fei Sun", "Mengnan Du" ]	[ "cs.LG", "cs.CL" ]	[ "Computer Science" ]	2026-06-08T00:00:00	https://arxiv.org/abs/2606.09043	https://arxiv.org/pdf/2606.09043v1	2606.09043	null	0	0	false	null	null	0.35
4c5eda60152941e6f8379b4a6d05fdafc731fd72c6158f78c4bcf5f7820190b9	[ "arxiv", "semantic_scholar" ]	When RLHF Fails: A Mechanistic Taxonomy of Reward Hacking, Collapse, and Evaluator Gaming	Reinforcement learning from human feedback (RLHF) makes large-scale post-training possible by replacing an underspecified human objective with learned and scalable proxies. The same substitution creates a structured failure surface: optimization can raise the learned reward while external quality falls, degrade both pr...	[ "Zelalem Abahana" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-06-02T00:00:00	https://arxiv.org/abs/2606.03238	https://arxiv.org/pdf/2606.03238v1	2606.03238	null	0	0	false	null	null	0.35
ffa7dd593613a170a8aefdf8639b244173ec134c9309d4131f852ba6e5d3818e	[ "arxiv", "semantic_scholar" ]	Sparse Mixture-of-Experts Reward Models Learn Interpretable and Specialized Experts for Personalized Preference Modeling	Preference modeling plays a central role in reinforcement learning from human feedback (RLHF), enabling large language models (LLMs) to align with human values. However, most existing approaches assume a universal reward function, neglecting the diversity and heterogeneity of human preferences. To address this limitati...	[ "Yifan Wang", "Jinyi Mu", "Mayank Jobanputra", "Yu Wang", "Ji-Ung Lee", "Soyoung Oh", "Isabel Valera", "Vera Demberg" ]	[ "cs.LG", "cs.AI", "cs.CL" ]	[ "Computer Science" ]	2026-06-02T00:00:00	https://arxiv.org/abs/2606.04284	https://arxiv.org/pdf/2606.04284v1	2606.04284	null	0	0	false	null	null	0.35
a8c63d3c04413eb395860ed58cbfd0470d17de452ab7a6ab9b740cdaeab7c469	[ "arxiv", "semantic_scholar" ]	EvalStop: Using World Feedback to Detect and Correct Reward Overoptimization in Multi-Tenant RLHF Platforms	Cloud LLM fine-tuning platforms increasingly serve RLHF workloads, where a learned reward model is optimized as a proxy for human quality. As Gao et al. (2023) showed, this proxy diverges from world feedback (downstream eval metrics) under sustained optimization pressure, a phenomenon known as reward overoptimization. ...	[ "Guilin Zhang", "Chuanyi Sun", "Kai Zhao", "Xu Chu", "Shahryar Sarkani", "John M. Fossaceca" ]	[ "cs.LG", "cs.AI", "cs.DC" ]	[ "Computer Science" ]	2026-06-02T00:00:00	https://arxiv.org/abs/2606.04145	https://arxiv.org/pdf/2606.04145v3	2606.04145	null	0	0	false	null	null	0.35
06b076868fb0a8d103d7e1af2f5e84f669e7dfc9f50ea242f44e20bda9b83d09	[ "arxiv", "semantic_scholar" ]	RDA: Reward Design Agent for Reinforcement Learning	Reinforcement learning has enabled the acquisition of impressive robotic skills, but typically requires hand-crafted reward functions that are slow to design and difficult to align with human intentions. Recent work, such as Eureka, automates reward design by using an LLM to iteratively generate and refine reward code ...	[ "Hojoon Lee", "Ajay Subramanian", "Ben Abbatematteo", "Vijay Veerabadran", "Pedro Matias", "Karl Ridgeway", "Nitin Kamra" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-06-01T00:00:00	https://arxiv.org/abs/2606.01672	https://arxiv.org/pdf/2606.01672v1	2606.01672	null	0	0	false	null	null	0.35
40337a1a94e845a0063edbf9bc197dc27f9656b6d1fe319424d7d203a863045f	[ "arxiv", "semantic_scholar" ]	From Reward-Free Representations to Preferences: Rethinking Offline Preference-Based Reinforcement Learning	Preference-based reinforcement learning (PbRL) avoids explicit reward engineering by learning from pairwise human preference feedback. Existing offline PbRL methods typically follow a two-stage pipeline, first learning a reward or preference model from labeled preferences and then performing offline RL on unlabeled dat...	[ "Jun-Jie Yang", "Chia-Heng Hsu", "Kui-Yuan Chen", "Ping-Chun Hsieh" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-05-31T00:00:00	https://arxiv.org/abs/2606.01123	https://arxiv.org/pdf/2606.01123v1	2606.01123	null	0	0	true	https://github.com/rl-bandits-lab/FB-PbRL	null	0.65
8564bdf28b12160b90415ce403089eae6552e5a6ea6426e36e0391bb816d2151	[ "arxiv", "semantic_scholar" ]	The Representation-Rationalizability Tradeoff in Reward Learning	In RLHF, each training example contains a prompt $x$ and two candidate responses $y,y'$, and annotators provide pairwise preferences between these responses. The learning problem is to convert these heterogeneous pairwise judgments into a single scalar reward $r(x,y)$ that measures response quality for each prompt. Cla...	[ "Jing Dong", "Yaoliang Yu", "Pascal Pourpart" ]	[ "cs.GT", "cs.LG" ]	[ "Computer Science" ]	2026-05-29T00:00:00	https://arxiv.org/abs/2606.00291	https://arxiv.org/pdf/2606.00291v1	2606.00291	null	0	0	false	null	null	0.35
844ca3b2606f222e9163a7702cb4222d89f2a52c181a6567a2b70ab276ed2e20	[ "arxiv", "semantic_scholar" ]	In-Context Reward Adaptation for Robust Preference Modeling	Reinforcement Learning from Human Feedback (RLHF) typically relies on static reward models to align Large Language Models with human preferences. However, human values are inherently diverse and heterogeneous, and a single reward model often lacks the robustness required to generalize to unseen preference domains. Whil...	[ "Zhenyu Sun", "Zheng Xu", "Ermin Wei" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-05-28T00:00:00	https://arxiv.org/abs/2605.30323	https://arxiv.org/pdf/2605.30323v1	2605.30323	null	0	0	false	null	null	0.35
20fc28ab632906daf136f4c08b94f85d79a031ed75b149c41abdc4659973afe4	[ "arxiv", "semantic_scholar" ]	Reward Learning from Best-of-$N$ Preference Data: Targets, Tradeoffs, and Design Principles	Best-of-$N$ sampling is widely used to construct pairwise preference data: $N$ candidates are drawn from a base distribution, and the best is paired with a rejected response. Despite its widespread use, what Bradley--Terry (BT) reward learning extracts from such data, and how to choose $N$ and the base distribution, re...	[ "Rattana Pukdee", "Maria-Florina Balcan", "Pradeep Ravikumar" ]	[ "stat.ML", "cs.AI", "cs.LG" ]	[ "Mathematics", "Computer Science" ]	2026-05-28T00:00:00	https://arxiv.org/abs/2605.30619	https://arxiv.org/pdf/2605.30619v1	2605.30619	null	0	0	false	null	null	0.35
6e1f3104957ece7fbac9192701555b413b33bc2fb003f0935dfe2c6205ece5e7	[ "arxiv", "semantic_scholar" ]	Rubric-Guided Process Reward for Stepwise Model Routing	Stepwise model routing improves the efficiency of Large Reasoning Models (LRMs) by assigning each reasoning step to a suitable model. Recent methods formulate routing as a sequential decision process and train the router with reinforcement learning. However, although they model routing as a process, they still supervis...	[ "Shenghao Ye", "Yu Guo", "Zhengheng Li", "Shuangwu Chen", "Jian Yang" ]	[ "cs.AI", "cs.CL" ]	[ "Computer Science" ]	2026-05-28T00:00:00	https://arxiv.org/abs/2605.29310	https://arxiv.org/pdf/2605.29310v1	2605.29310	null	0	0	false	null	null	0.35
e8ccfdb70c48392eaa25d06ca579bf1115efef646d6885136194f0af89b5add6	[ "arxiv", "semantic_scholar" ]	Beyond Pairwise Preferences: Listwise Reward-Aware Alignment for Diffusion Models	Preference optimization has emerged as an efficient alternative to online reinforcement learning from human feedback (RLHF) for aligning text-to-image diffusion models. However, existing methods largely reduce supervision to binary pairwise comparisons. This pairwise reduction is limiting when training data naturally c...	[ "Austin Wang", "Jiaqi Han", "Stefano Ermon", "Yisong Yue" ]	[ "cs.LG", "cs.CV" ]	[ "Computer Science" ]	2026-05-26T00:00:00	https://arxiv.org/abs/2605.26491	https://arxiv.org/pdf/2605.26491v1	2605.26491	null	0	0	false	null	null	0.35
7d765b8ec398948d46ce164bc9fde1a1ae580462a14d12d4fedbc96026cde05e	[ "arxiv", "semantic_scholar" ]	Focal Reward: Balanced Reinforcement Learning under Rubric-Based Rewards	The open-ended generation in LLMs usually requires multi-dimensional rubrics to adequately assess quality and guide the improvement of reinforcement learning. However, a critical dilemma inherent in this training paradigm is the imbalanced reward polarization along different rubric dimensions. Under this bottleneck, ev...	[ "Yu Huang", "Zihua Zhao", "Zhaoxin Huan", "Wanli Gu", "Feng Hong", "Xinmu Ge", "Lin Yuan", "Weichang Wu", "Qiang Hu", "Xiaolu Zhang", "Jun Zhou", "Jiangchao Yao" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-05-26T00:00:00	https://arxiv.org/abs/2605.26579	https://arxiv.org/pdf/2605.26579v1	2605.26579	null	1	0	false	null	null	0.35
5ddbc515851e6d77372c247310dfb30c3093b4284e663b00c667ba4e393eeb06	[ "arxiv", "semantic_scholar" ]	How Neural Reward Models Learn Features for Policy Optimization: A Single-Index Analysis	Reward modeling is not only a prediction problem: in KL-regularized policy optimization, the learned reward is exponentiated to define the deployed policy, so downstream value depends on errors in reward-tilted regions. We study this feedback in a Gaussian single-index model with $r^(x) = σ^(\langle θ^*, x\rangle)$ a...	[ "Rei Higuchi", "Ryotaro Kawata", "Akifumi Wachi", "Shokichi Takakura", "Kohei Miyaguchi", "Taiji Suzuki" ]	[ "stat.ML", "cs.LG" ]	[ "Mathematics", "Computer Science" ]	2026-05-23T00:00:00	https://arxiv.org/abs/2605.24749	https://arxiv.org/pdf/2605.24749v1	2605.24749	null	0	0	false	null	null	0.35
ab81af1d9787cef06f5c03c73c7bd8eb042707b2870328d761aa8d02e024b6e6	[ "arxiv", "semantic_scholar" ]	PREFINE: Preference-Based Implicit Reward and Cost Fine-Tuning for Safety Alignment	We address the problem of making a pre-trained reinforcement learning (RL) policy safety-aware by incorporating cost constraints without retraining it from scratch. While costs could be numerically encoded, we assume a more general setting is when costs are provided as preferences. Given a reward-optimized policy and a...	[ "Richa Verma", "Bavish Kulur", "Sanjay Chawla", "Balaraman Ravindran" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-05-20T00:00:00	https://arxiv.org/abs/2605.21225	https://arxiv.org/pdf/2605.21225v1	2605.21225	10.65109/sdrb4374	0	0	false	null	null	0.35
c74e533a4ffd3bbd16f4460f50dab708a3da199f34d89d8c929269da45cf0869	[ "arxiv", "semantic_scholar" ]	Process Rewards with Learned Reliability	Process Reward Models (PRMs) provide step-level feedback for reasoning, but current PRMs usually output only a single reward score for each step. Downstream methods must therefore treat imperfect step-level reward predictions as reliable decision signals, with no indication of when these predictions should be trusted. ...	[ "Jinyuan Li", "Langlin Huang", "Chengsong Huang", "Shaoyang Xu", "Donghong Cai", "Yuyi Yang", "Wenxuan Zhang", "Jiaxin Huang" ]	[ "cs.CL", "cs.AI", "cs.LG" ]	[ "Computer Science" ]	2026-05-15T00:00:00	https://arxiv.org/abs/2605.15529	https://arxiv.org/pdf/2605.15529v1	2605.15529	null	0	0	false	null	null	0.35
a724fc1423e57ea3892080898e88cd92924d6946340a2a62e1194d460e6a6b9a	[ "arxiv", "semantic_scholar" ]	Variance-aware Reward Modeling with Anchor Guidance	Standard Bradley--Terry (BT) reward models are limited when human preferences are pluralistic. Although soft preference labels preserve disagreement information, BT can only express it by shrinking reward margins. Gaussian reward models provide an alternative by jointly predicting a reward mean and a reward variance, b...	[ "Shuxing Fang", "Ruijian Han", "Liangyu Zhang", "Fan Zhou" ]	[ "stat.ML", "cs.LG" ]	[ "Mathematics", "Computer Science" ]	2026-05-12T00:00:00	https://arxiv.org/abs/2605.11865	https://arxiv.org/pdf/2605.11865v1	2605.11865	null	0	0	false	null	null	0.35
18e939e0a31d45dbe0a09aabd7f95f26cbc0f600567ba040b98259bb6f97c109	[ "arxiv", "semantic_scholar" ]	Unsupervised Process Reward Models	Process Reward Models (PRMs) are a powerful mechanism for steering large language model reasoning by providing fine-grained, step-level supervision. However, this effectiveness comes at a significant cost: PRMs require expert annotations for every reasoning step, making them costly and difficult to scale. Here, we prop...	[ "Artyom Gadetsky", "Maxim Kodryan", "Siba Smarak Panigrahi", "Hang Guo", "Maria Brbic" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-05-11T00:00:00	https://arxiv.org/abs/2605.10158	https://arxiv.org/pdf/2605.10158v1	2605.10158	null	0	0	false	null	null	0.35
065489ca5e818afb74c9b2672e38d9cf7e575bb2e927e5945881d4a9f3916acb	[ "arxiv", "semantic_scholar" ]	Spurious Correlation Learning in Preference Optimization: Mechanisms, Consequences, and Mitigation via Tie Training	Preference learning methods like Direct Preference Optimization (DPO) are known to induce reliance on spurious correlations, leading to sycophancy and length bias in today's language models and potentially severe goal misgeneralization in future systems. In this work, we provide a unified theoretical analysis of this p...	[ "Christian Moya", "Alex Semendinger", "Guang Lin", "Elliott Thornley" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-05-11T00:00:00	https://arxiv.org/abs/2605.11134	https://arxiv.org/pdf/2605.11134v2	2605.11134	null	0	0	false	null	Proceedings of the 43rd International Conference on Machine Learning, 2026, Seoul, South Korea	0.55
b73983bba8a2b09656b1a0b2df466bdce3688205c4af0f8b4a1ac0f58b959a5e	[ "arxiv", "semantic_scholar" ]	$ξ$-DPO: Direct Preference Optimization via Ratio Reward Margin	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 hyperparamete...	[ "Zhengyuan Fan", "Zhonghua Wu", "Yuxuan Du", "Qun Chen" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-05-09T00:00:00	https://arxiv.org/abs/2605.10981	https://arxiv.org/pdf/2605.10981v1	2605.10981	null	0	0	false	null	null	0.35
4033de934a2c65fc0a94e7348ed3813ecfd866e349786c733922fe1989038e90	[ "arxiv", "semantic_scholar" ]	Preference Instability in Reward Models: Detection and Mitigation via Sparse Autoencoders	Preference learning in large language models relies on reward models as proxies for human judgment. However, these models frequently exhibit preference instability, producing contradictory preference assignments in response to subtle, meaning-preserving input variations. We analyze this instability at the representatio...	[ "Shunchang Liu", "Xin Chen", "Belen Martin Urcelay", "Francesco Croce" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-05-07T00:00:00	https://arxiv.org/abs/2605.16339	https://arxiv.org/pdf/2605.16339v1	2605.16339	null	0	0	true	https://github.com/shunchang-liu/pisa}	null	0.65
aeb37bbf2614e2a8cda00646ba389d0c1c23824cd3b7bade15b0ab05fd22afd6	[ "arxiv", "semantic_scholar" ]	Optimal Transport for LLM Reward Modeling from Noisy Preference	Reward models are fundamental to Reinforcement Learning from Human Feedback (RLHF), yet real-world datasets are inevitably corrupted by noisy preference. Conventional training objectives tend to overfit these errors, while existing denoising approaches often rely on homogeneous noise assumptions that fail to capture th...	[ "Licheng Pan", "Haochen Yang", "Haoxuan Li", "Yunsheng Lu", "Yongqi Tong", "Yinuo Wang", "Shijian Wang", "Zhixuan Chu", "Lei Shen", "Yuan Lu", "Hao Wang" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-05-07T00:00:00	https://arxiv.org/abs/2605.06036	https://arxiv.org/pdf/2605.06036v1	2605.06036	null	0	0	false	null	null	0.35
dddc48c0563e8fc55c53d1e2008fbd2ade4997d8d23f6d404bde57e11b365b82	[ "arxiv", "semantic_scholar" ]	Misaligned by Reward: Socially Undesirable Preferences in LLMs	Reward models are a key component of large language model alignment, serving as proxies for human preferences during training. However, existing evaluations focus primarily on broad instruction-following benchmarks, providing limited insight into whether these models capture socially desirable preferences. As a result,...	[ "Gayane Ghazaryan", "Esra Dönmez" ]	[ "cs.CL", "cs.AI", "cs.CY" ]	[ "Computer Science" ]	2026-05-06T00:00:00	https://arxiv.org/abs/2605.05003	https://arxiv.org/pdf/2605.05003v1	2605.05003	null	0	0	false	null	null	0.35
dd9ac75204f67cadf4b7605710e0e54ed8127f921bd72c7fbdbc8dd96e5c7adc	[ "arxiv", "semantic_scholar" ]	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 r...	[ "Yangyang Zhou", "Yi-Chen Li" ]	[ "cs.CL", "cs.AI" ]	[ "Computer Science" ]	2026-05-03T00:00:00	https://arxiv.org/abs/2605.01831	https://arxiv.org/pdf/2605.01831v1	2605.01831	null	0	0	true	https://github.com/nanzhi84/RMGAP	null	0.65
a175076b1e4da2c8f0bfdc8bf45d60fcdd7a9353faa5081acef1c7f2da9f0146	[ "arxiv", "semantic_scholar" ]	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 partial...	[ "Ziqin Yuan", "Ruiqi Wang", "Dezhong Zhao", "Baijian Yang", "Byung-Cheol Min" ]	[ "cs.RO" ]	[ "Computer Science" ]	2026-05-01T00:00:00	https://arxiv.org/abs/2605.00384	https://arxiv.org/pdf/2605.00384v1	2605.00384	null	0	0	false	null	null	0.35
bacd73bd7af44419d88f197e663d70e8b86fdf93f365435948ea2bb4e927c076	[ "arxiv", "semantic_scholar" ]	Uncertainty-Aware Reward Discounting for Mitigating Reward Hacking	Reinforcement learning (RL) systems typically optimize scalar reward functions that assume precise and reliable evaluation of outcomes. However, real-world objectives--especially those derived from human preferences--are often uncertain, context-dependent, and internally inconsistent. This mismatch can lead to alignmen...	[ "Disha Singha" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-04-29T00:00:00	https://arxiv.org/abs/2604.26360	https://arxiv.org/pdf/2604.26360v1	2604.26360	10.48550/arXiv.2604.26360	0	0	false	null	arXiv.org	0.55
9c20d095227a499cdf65ee8b3ff8e2716a9a92d2d07e4bb37d4efef9b2305b6b	[ "arxiv", "semantic_scholar" ]	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 sca...	[ "Mohammed Suhail B Nadaf" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-04-28T00:00:00	https://arxiv.org/abs/2604.26130	https://arxiv.org/pdf/2604.26130v1	2604.26130	10.48550/arXiv.2604.26130	1	0	true	https://github.com/suhailnadaf509/reward-lens	arXiv.org	0.85
377546b10133d0877254ba3ade1fc5ec3e16068b9e9d8543a42d76bf986005f3	[ "arxiv", "semantic_scholar" ]	Rewarding the Scientific Process: Process-Level Reward Modeling for Agentic Data Analysis	Process Reward Models (PRMs) have achieved remarkable success in augmenting the reasoning capabilities of Large Language Models (LLMs) within static domains such as mathematics. However, their potential in dynamic data analysis tasks remains underexplored. In this work, we first present a empirical study revealing that...	[ "Zhisong Qiu", "Shuofei Qiao", "Kewei Xu", "Yuqi Zhu", "Lun Du", "Ningyu Zhang", "Huajun Chen" ]	[ "cs.CL", "cs.AI", "cs.CE", "cs.LG", "cs.MA" ]	[ "Computer Science" ]	2026-04-27T00:00:00	https://arxiv.org/abs/2604.24198	https://arxiv.org/pdf/2604.24198v1	2604.24198	10.48550/arXiv.2604.24198	3	0	true	https://github.com/zjunlp/DataMind	arXiv.org	0.85
927c290a876589751e7f70d291e3b0fa85230a853ad3f9e6a605682bb7df10af	[ "arxiv", "semantic_scholar" ]	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 condi...	[ "Alex Nikulkov" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-04-24T00:00:00	https://arxiv.org/abs/2604.22981	https://arxiv.org/pdf/2604.22981v1	2604.22981	10.48550/arXiv.2604.22981	1	0	false	null	arXiv.org	0.55
0b523a07c46024bc53dcbeca2dc03d867eb2b77f0a0c8f9b84c0dac1b8d1575b	[ "arxiv", "semantic_scholar" ]	K-Score: Kalman Filter as a Principled Alternative to Reward Normalization in Reinforcement Learning	We propose a simple yet effective alternative to reward normalization in policy gradient reinforcement learning by integrating a 1D Kalman filter for online reward estimation. Instead of relying on fixed heuristics, our method recursively estimates the latent reward mean, smoothing high-variance returns and adapting to...	[ "Zixuan Xia", "Quanxi Li" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-04-24T00:00:00	https://arxiv.org/abs/2604.23056	https://arxiv.org/pdf/2604.23056v1	2604.23056	10.48550/arXiv.2604.23056	0	0	true	https://github.com/Sumxiaa/Kalman_Normalization	arXiv.org	0.85
e750fa7fc69af813791eadc3212ebeb6415c72866aaf6181ca116f1976264c68	[ "arxiv", "semantic_scholar" ]	Mitigating Multimodal Hallucination via Phase-wise Self-reward	Large Vision-Language Models (LVLMs) still struggle with vision hallucination, where generated responses are inconsistent with the visual input. Existing methods either rely on large-scale annotated data for fine-tuning, which incurs massive computational overhead, or employ static post-hoc strategies that overlook the...	[ "Yu Zhang", "Chuyang Sun", "Kehai Chen", "Xuefeng Bai", "Yang Xiang", "Min Zhang" ]	[ "cs.CV", "cs.CL" ]	[ "Computer Science" ]	2026-04-20T00:00:00	https://arxiv.org/abs/2604.17982	https://arxiv.org/pdf/2604.17982v1	2604.17982	10.48550/arXiv.2604.17982	2	0	false	null	arXiv.org	0.55
488df9c33414a110b0eabcf74d54efec3f5de9688c65b631c60318667c9a5027	[ "arxiv", "semantic_scholar" ]	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 underexp...	[ "Xingyu Fan", "Wei Shao", "Jiacheng Liu", "Linqi Song", "Pheng Ann Heng" ]	[ "cs.AI", "cs.CL" ]	[ "Computer Science" ]	2026-04-20T00:00:00	https://arxiv.org/abs/2604.18327	https://arxiv.org/pdf/2604.18327v1	2604.18327	10.1109/jstsp.2026.3690098	0	0	false	null	IEEE Journal on Selected Topics in Signal Processing	0.55
5c49c09021c7d5ab0e7d77041e89345ccd7a3b431a2493251e6ab48b9d2a151f	[ "arxiv", "semantic_scholar" ]	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 cooperatio...	[ "Akira Kawabata", "Saku Sugawara" ]	[ "cs.CL", "cs.LG" ]	[ "Computer Science" ]	2026-04-15T00:00:00	https://arxiv.org/abs/2604.13618	https://arxiv.org/pdf/2604.13618v1	2604.13618	10.48550/arXiv.2604.13618	1	0	false	null	arXiv.org	0.5443
e6e74114ac20ae26049b171efcb6f3003d69530623b6d38241c7d727af695027	[ "arxiv", "semantic_scholar" ]	Enhancing Reinforcement Learning for Radiology Report Generation with Evidence-aware Rewards and Self-correcting Preference Learning	Recent reinforcement learning (RL) approaches have advanced radiology report generation (RRG), yet two core limitations persist: (1) report-level rewards offer limited evidence-grounded guidance for clinical faithfulness; and (2) current methods lack an explicit self-improving mechanism to align with clinical preferenc...	[ "Qin Zhou", "Guoyan Liang", "Qianyi Yang", "Jingyuan Chen", "Sai Wu", "Chang Yao", "Zhe Wang" ]	[ "cs.LG", "stat.ME" ]	[ "Computer Science", "Mathematics" ]	2026-04-15T00:00:00	https://arxiv.org/abs/2604.13598	https://arxiv.org/pdf/2604.13598v1	2604.13598	10.48550/arXiv.2604.13598	0	0	false	null	arXiv.org	0.5443
70528fd8625a16e32b22bbb1e4e3f68891e0c39f400175acae7b0582a8805852	[ "arxiv", "semantic_scholar" ]	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. U...	[ "Tiantian Zhang", "Jierui Zuo", "Michael Chen", "Wenping Wang" ]	[ "stat.ML", "cs.LG" ]	[ "Computer Science", "Mathematics" ]	2026-04-13T00:00:00	https://arxiv.org/abs/2604.11119	https://arxiv.org/pdf/2604.11119v2	2604.11119	10.48550/arXiv.2604.11119	0	0	false	null	arXiv.org	0.542
16b03ee8ce722b99ee0a46924915bfb54c790ff7296d08b3fecdbe2159c7dea8	[ "arxiv", "semantic_scholar" ]	Mitigating Reward Hacking in RLHF via Advantage Sign Robustness	Reward models (RMs) used in reinforcement learning from human feedback (RLHF) are vulnerable to reward hacking: as the policy maximizes a learned proxy reward, true quality plateaus or degrades. We make the assumption that reward hacking is often caused by flipped advantage signs: instead of reducing the likelihood of ...	[ "Shinnosuke Ono", "Johannes Ackermann", "Soichiro Nishimori", "Takashi Ishida", "Masashi Sugiyama" ]	[ "cs.LG", "cs.AI", "cs.CL" ]	[ "Computer Science" ]	2026-04-03T00:00:00	https://arxiv.org/abs/2604.02986	https://arxiv.org/pdf/2604.02986v1	2604.02986	10.48550/arXiv.2604.02986	1	0	false	null	arXiv.org	0.5305
3531a9a3b952f22acbd55e51d7eb154c37648c0905d56eea4ed2a553fd4755c0	[ "arxiv", "semantic_scholar" ]	PAC-Bayesian Reward-Certified Outcome Weighted Learning	Estimating optimal individualized treatment rules (ITRs) via outcome weighted learning (OWL) often relies on observed rewards that are noisy or optimistic proxies for the true latent utility. Ignoring this reward uncertainty leads to the selection of policies with inflated apparent performance, yet existing OWL framewo...	[ "Yuya Ishikawa", "Shu Tamano" ]	[ "cs.LG", "stat.ME", "stat.ML" ]	[ "Computer Science", "Mathematics" ]	2026-04-02T00:00:00	https://arxiv.org/abs/2604.01946	https://arxiv.org/pdf/2604.01946v1	2604.01946	10.48550/arXiv.2604.01946	0	0	false	null	arXiv.org	0.5294
2c74fde5a11989ebf10a786fa84414f456c829735004804bedbb79fbd43a5c6e	[ "arxiv", "semantic_scholar" ]	Preference learning in shades of gray: Interpretable and bias-aware reward modeling for human preferences	Learning human preferences in language models remains fundamentally challenging, as reward modeling relies on subtle, subjective comparisons or shades of gray rather than clear-cut labels. This study investigates the limits of current approaches and proposes a feature-augmented framework to better capture the multidime...	[ "Simona-Vasilica Oprea", "Adela Bâra" ]	[ "cs.CL", "cs.AI" ]	[ "Computer Science" ]	2026-04-01T00:00:00	https://arxiv.org/abs/2604.01312	https://arxiv.org/pdf/2604.01312v1	2604.01312	10.48550/arXiv.2604.01312	0	0	false	null	arXiv.org	0.5282
d55bdbfe95a58da4928d8d5dd6786c6ca701f36332954ef6a0076eb23830074e	[ "arxiv", "semantic_scholar" ]	ImplicitRM: Unbiased Reward Modeling from Implicit Preference Data for LLM alignment	Reward modeling represents a long-standing challenge in reinforcement learning from human feedback (RLHF) for aligning language models. Current reward modeling is heavily contingent upon experimental feedback data with high collection costs. In this work, we study \textit{implicit reward modeling} -- learning reward mo...	[ "Hao Wang", "Haocheng Yang", "Licheng Pan", "Lei Shen", "Xiaoxi Li", "Yinuo Wang", "Zhichao Chen", "Yuan Lu", "Haoxuan Li", "Zhouchen Lin" ]	[ "cs.CL", "cs.AI", "stat.AP" ]	[ "Computer Science", "Mathematics" ]	2026-03-24T00:00:00	https://arxiv.org/abs/2603.23184	https://arxiv.org/pdf/2603.23184v1	2603.23184	10.48550/arXiv.2603.23184	0	0	true	null	arXiv.org	0.8022
6f6a1eaef745956cf726a27aac38d02bc26068dd2fb7012368029ea1781768eb	[ "arxiv", "semantic_scholar" ]	Privacy-Preserving Reinforcement Learning from Human Feedback via Decoupled Reward Modeling	Preference-based fine-tuning has become an important component in training large language models, and the data used at this stage may contain sensitive user information. A central question is how to design a differentially private pipeline that is well suited to the distinct structure of reinforcement learning from hum...	[ "Young Hyun Cho", "Will Wei Sun" ]	[ "stat.ML", "cs.LG" ]	[ "Computer Science", "Mathematics" ]	2026-03-23T00:00:00	https://arxiv.org/abs/2603.22563	https://arxiv.org/pdf/2603.22563v1	2603.22563	10.48550/arXiv.2603.22563	3	0	false	null	arXiv.org	0.5179
04ba9ecb076ea88060dd8beac0e4df9f84ceda7cedf0cff73be6d3d24c05c7d0	[ "arxiv", "semantic_scholar" ]	Reward Sharpness-Aware Fine-Tuning for Diffusion Models	Reinforcement learning from human feedback (RLHF) has proven effective in aligning large language models with human preferences, inspiring the development of reward-centric diffusion reinforcement learning (RDRL) to achieve similar alignment and controllability. While diffusion models can generate high-quality outputs,...	[ "Kwanyoung Kim", "Byeongsu Sim" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-03-22T00:00:00	https://arxiv.org/abs/2603.21175	https://arxiv.org/pdf/2603.21175v1	2603.21175	10.48550/arXiv.2603.21175	0	0	false	null	arXiv.org	0.5168
b7755cda26a08d6a0a98515a4029778a35d0c5354bd55321b4cd2f6d9c77f4d1	[ "arxiv", "semantic_scholar" ]	CausalRM: Causal-Theoretic Reward Modeling for RLHF from Observational User Feedbacks	Despite the success of reinforcement learning from human feedback (RLHF) in aligning language models, current reward modeling heavily relies on experimental feedback data collected from human annotators under controlled and costly conditions. In this work, we introduce observational reward modeling -- learning reward m...	[ "Hao Wang", "Licheng Pan", "Zhichao Chen", "Chunyuan Zheng", "Zhixuan Chu", "Xiaoxi Li", "Yuan Lu", "Xinggao Liu", "Haoxuan Li", "Zhouchen Lin" ]	[ "cs.LG", "cs.AI", "cs.CL", "stat.ML" ]	[ "Computer Science", "Mathematics" ]	2026-03-19T00:00:00	https://arxiv.org/abs/2603.18736	https://arxiv.org/pdf/2603.18736v1	2603.18736	10.48550/arXiv.2603.18736	1	0	true	null	arXiv.org	0.7933
81a02f3ea5e263070c212c956e961ab23ba89b52a58069ec9d8bb2b7f2027cae	[ "arxiv", "semantic_scholar" ]	Robust Post-Training for Generative Recommenders: Why Exponential Reward-Weighted SFT Outperforms RLHF	Aligning generative recommender systems to user preferences via post-training is critical for closing the gap between next-item prediction and actual recommendation quality. Existing post-training methods are ill-suited for production-scale systems: RLHF methods reward hack due to noisy user feedback and unreliable rew...	[ "Keertana Chidambaram", "Sanath Kumar Krishnamurthy", "Qiuling Xu", "Ko-Jen Hsiao", "Moumita Bhattacharya" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-03-10T00:00:00	https://arxiv.org/abs/2603.10279	https://arxiv.org/pdf/2603.10279v1	2603.10279	10.48550/arXiv.2603.10279	0	0	true	null	arXiv.org	0.7774
4219af8f68a24e0e4bf67a43c1224bcd7f0313b8860afb03c7427c02a32a27fa	[ "arxiv", "semantic_scholar" ]	Causally Robust Reward Learning from Reason-Augmented Preference Feedback	Preference-based reward learning is widely used for shaping agent behavior to match a user's preference, yet its sparse binary feedback makes it especially vulnerable to causal confusion. The learned reward often latches onto spurious features that merely co-occur with preferred trajectories during training, collapsing...	[ "Minjune Hwang", "Yigit Korkmaz", "Daniel Seita", "Erdem Bıyık" ]	[ "cs.AI", "cs.LG", "cs.RO" ]	[ "Computer Science" ]	2026-03-05T00:00:00	https://arxiv.org/abs/2603.04861	https://arxiv.org/pdf/2603.04861v1	2603.04861	10.48550/arXiv.2603.04861	0	0	true	https://github.com/mj-hwang/ReCouPLe	arXiv.org	0.7685
cce5dccf585671458d0f4fd545b84948f2cf4e2cf04b345d1f7b4ff00cb5013f	[ "arxiv", "semantic_scholar" ]	VRM: Teaching Reward Models to Understand Authentic Human Preferences	Large Language Models (LLMs) have achieved remarkable success across diverse natural language tasks, yet the reward models employed for aligning LLMs often encounter challenges of reward hacking, where the approaches predominantly rely on directly mapping prompt-response pairs to scalar scores, which may inadvertently ...	[ "Biao Liu", "Ning Xu", "Junming Yang", "Hao Xu", "Xin Geng" ]	[ "cs.CL" ]	[ "Computer Science" ]	2026-03-05T00:00:00	https://arxiv.org/abs/2603.04974	https://arxiv.org/pdf/2603.04974v1	2603.04974	10.48550/arXiv.2603.04974	0	0	false	null	arXiv.org	0.4973
3064227dd486b1b97e8bb41ab984bd7db9db8840a0d2c571e27e0835be210f96	[ "arxiv", "semantic_scholar" ]	Reward-Conditioned Reinforcement Learning	Single-task RL agents are typically trained under a fixed reward function, which limits their robustness to reward misspecification and their ability to adapt to changing preferences. We introduce Reward-Conditioned Reinforcement Learning (RCRL), an off-policy method that conditions agents on reward parameterizations w...	[ "Michal Nauman", "Marek Cygan", "Pieter Abbeel" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-03-05T00:00:00	https://arxiv.org/abs/2603.05066	https://arxiv.org/pdf/2603.05066v3	2603.05066	10.48550/arXiv.2603.05066	0	0	false	null	arXiv.org	0.4973
505aff0dcfb0d2d18e2f487adb8f45a371efe9008837e2b55dbb6b690656042e	[ "arxiv", "semantic_scholar" ]	Alternating Reinforcement Learning with Contextual Rubric Rewards: Beyond the Scalarization Strategy	Reinforcement Learning with Rubric Rewards (RLRR) is a framework that extends conventional reinforcement learning from human feedback (RLHF) and verifiable rewards (RLVR) by replacing scalar preference signals with structured, multi-dimensional, contextual rubric-based evaluations. However, existing approaches in RLRR ...	[ "Guangchen Lan", "Lian Xiong", "Xin Zhou", "Hejie Cui", "Yuwei Zhang", "Mao Li", "Zhenyu Shi", "Besnik Fetahu", "Lihong Li", "Xian Li" ]	[ "cs.LG", "cs.AI", "cs.CL" ]	[ "Computer Science" ]	2026-03-04T00:00:00	https://arxiv.org/abs/2603.15646	https://arxiv.org/pdf/2603.15646v2	2603.15646	null	1	0	false	null	null	0.3157
d6a0a3b1cce8eeccf718a8b38a227fd78986f08c14b14ed109873e77cf55dbea	[ "arxiv", "semantic_scholar" ]	Generalisation of RLHF under Reward Shift and Clipped KL Regularisation	Alignment and adaptation in large language models heavily rely on reinforcement learning from human feedback (RLHF); yet, theoretical understanding of its generalisability remains premature, especially when the learned reward could shift, and the KL control is estimated and clipped. To address this issue, we develop ge...	[ "Kenton Tang", "Yuzhu Chen", "Fengxiang He" ]	[ "cs.LG", "cs.AI", "stat.ML" ]	[ "Computer Science", "Mathematics" ]	2026-02-25T00:00:00	https://arxiv.org/abs/2602.21765	https://arxiv.org/pdf/2602.21765v1	2602.21765	10.48550/arXiv.2602.21765	0	0	false	null	arXiv.org	0.4881
e37d453a9265820365d533c683bd4c97821ea0ed6eb7c87d3b748451d57929a7	[ "arxiv", "semantic_scholar" ]	The Art of Efficient Reasoning: Data, Reward, and Optimization	Large Language Models (LLMs) consistently benefit from scaled Chain-of-Thought (CoT) reasoning, but also suffer from heavy computational overhead. To address this issue, efficient reasoning aims to incentivize short yet accurate thinking trajectories, typically through reward shaping with Reinforcement Learning (RL). I...	[ "Taiqiang Wu", "Zenan Xu", "Bo Zhou", "Ngai Wong" ]	[ "cs.CL", "cs.AI" ]	[ "Computer Science" ]	2026-02-24T00:00:00	https://arxiv.org/abs/2602.20945	https://arxiv.org/pdf/2602.20945v3	2602.20945	10.48550/arXiv.2602.20945	2	0	false	null	arXiv.org	0.487
33167b1d802e267f69577dc16b98db9db2e288d66559296187d9cbbcf0a686ed	[ "arxiv", "semantic_scholar" ]	IR$^3$: Contrastive Inverse Reinforcement Learning for Interpretable Detection and Mitigation of Reward Hacking	Reinforcement Learning from Human Feedback (RLHF) enables powerful LLM alignment but can introduce reward hacking - models exploit spurious correlations in proxy rewards without genuine alignment. Compounding this, the objectives internalized during RLHF remain opaque, making hacking behaviors difficult to detect or co...	[ "Mohammad Beigi", "Ming Jin", "Junshan Zhang", "Jiaxin Zhang", "Qifan Wang", "Lifu Huang" ]	[ "cs.AI", "cs.LG" ]	[ "Computer Science" ]	2026-02-23T00:00:00	https://arxiv.org/abs/2602.19416	https://arxiv.org/pdf/2602.19416v1	2602.19416	10.48550/arXiv.2602.19416	3	0	false	null	arXiv.org	0.4858
a7f61a901a3e4b2026e4f73c945d6927b84a80ea4fb05588971d257cbfaf9fe9	[ "arxiv", "semantic_scholar" ]	Gradient Regularization Prevents Reward Hacking in Reinforcement Learning from Human Feedback and Verifiable Rewards	Reinforcement Learning from Human Feedback (RLHF) or Verifiable Rewards (RLVR) are two key steps in the post-training of modern Language Models (LMs). A common problem is reward hacking, where the policy may exploit inaccuracies of the reward and learn an unintended behavior. Most previous works address this by limitin...	[ "Johannes Ackermann", "Michael Noukhovitch", "Takashi Ishida", "Masashi Sugiyama" ]	[ "cs.LG", "cs.AI", "cs.CL" ]	[ "Computer Science" ]	2026-02-20T00:00:00	https://arxiv.org/abs/2602.18037	https://arxiv.org/pdf/2602.18037v1	2602.18037	10.48550/arXiv.2602.18037	4	0	false	null	arXiv.org	0.4824
67dcf1b604136a621eb4fa56229acea84ff4ea6f4fbc42cabaaa8e1ec55a6d5a	[ "arxiv", "semantic_scholar" ]	Reward Under Attack: Analyzing the Robustness and Hackability of Process Reward Models	Process Reward Models (PRMs) are rapidly becoming the backbone of LLM reasoning pipelines, yet we demonstrate that state-of-the-art PRMs are systematically exploitable under adversarial optimization pressure. To address this, we introduce a three-tiered diagnostic framework that applies increasing adversarial pressure ...	[ "Rishabh Tiwari", "Aditya Tomar", "Udbhav Bamba", "Monishwaran Maheswaran", "Heng Yang", "Michael W. Mahoney", "Kurt Keutzer", "Amir Gholami" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-02-20T00:00:00	https://arxiv.org/abs/2603.06621	https://arxiv.org/pdf/2603.06621v1	2603.06621	10.48550/arXiv.2603.06621	2	0	true	https://github.com/SqueezeAILab/reward-under-attack	arXiv.org	0.7455
86b41c50c6a849d49e1748be05dc3ba199cbcfde880e33334135121ede283f37	[ "arxiv", "semantic_scholar" ]	MARS: Margin and Semantic-Aware Data Augmentation for Reward Modeling	Reward modeling is central to alignment pipelines such as RLHF, RLAIF, and PPO-based policy optimization, yet its reliability is constrained by limited and heterogeneous human preference data that are expensive to collect at scale. While synthetic augmentation can expand preference supervision, existing methods often a...	[ "Payel Bhattacharjee", "Osvaldo Simeone", "Ravi Tandon" ]	[ "cs.LG", "cs.AI", "cs.IT" ]	[ "Computer Science", "Mathematics" ]	2026-02-19T00:00:00	https://arxiv.org/abs/2602.17658	https://arxiv.org/pdf/2602.17658v2	2602.17658	null	0	0	false	null	null	0.3062
93dd90acd3e8e82783f0433ea7131f25dc0584e13f079e198ace2776a57f6b6d	[ "arxiv", "semantic_scholar" ]	Beyond Binary Preferences: A Principled Framework for Reward Modeling with Ordinal Feedback	Reward modeling is crucial for aligning large language models with human preferences, yet current approaches lack a principled mathematical framework for leveraging ordinal preference data. When human annotators provide graded preferences on a Likert scale (e.g., significantly better, better, slightly better, negligibl...	[ "Amirhossein Afsharrad", "Ruida Zhou", "Luca Viano", "Sanjay Lall", "Mohammad Ghavamzadeh" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-02-13T00:00:00	https://arxiv.org/abs/2603.02232	https://arxiv.org/pdf/2603.02232v1	2603.02232	10.48550/arXiv.2603.02232	2	0	false	null	arXiv.org	0.4744
6fca7a4418a93f8d902956a2a6d9644a190a1df0a5054e883259700a30996269	[ "arxiv", "semantic_scholar" ]	Mitigating Reward Hacking in RLHF via Bayesian Non-negative Reward Modeling	Reward models learned from human preferences are central to aligning large language models (LLMs) via reinforcement learning from human feedback, yet they are often vulnerable to reward hacking due to noisy annotations and systematic biases such as response length or style. We propose Bayesian Non-Negative Reward Model...	[ "Zhibin Duan", "Guowei Rong", "Zhuo Li", "Bo Chen", "Mingyuan Zhou", "Dandan Guo" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-02-11T00:00:00	https://arxiv.org/abs/2602.10623	https://arxiv.org/pdf/2602.10623v2	2602.10623	10.48550/arXiv.2602.10623	2	0	true	https://github.com/GuoweiRong/Bayesian-Non-negative-Reward-Model	arXiv.org	0.7296
c7a50c90ecb99ef8c56cee427a936ace5443babf80db584a72b4b6dbe7531a36	[ "arxiv", "semantic_scholar" ]	Reward Modeling for Reinforcement Learning-Based LLM Reasoning: Design, Challenges, and Evaluation	Large Language Models (LLMs) demonstrate transformative potential, yet their reasoning remains inconsistent and unreliable. Reinforcement learning (RL)-based fine-tuning is a key mechanism for improvement, but its effectiveness is fundamentally governed by reward design. Despite its importance, the relationship between...	[ "Pei-Chi Pan", "Yingbin Liang", "Sen Lin" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-02-10T00:00:00	https://arxiv.org/abs/2602.09305	https://arxiv.org/pdf/2602.09305v1	2602.09305	10.48550/arXiv.2602.09305	3	0	false	null	arXiv.org	0.4709
daf167b4a8054a79bd16fae3e332d9df12e9ca647dbb265613b596351fe274cf	[ "arxiv", "semantic_scholar" ]	Bayesian Preference Learning for Test-Time Steerable Reward Models	Reward models are central to aligning language models with human preferences via reinforcement learning (RL). As RL is increasingly applied to settings such as verifiable rewards and multi-objective alignment, RMs are expected to encode more complex and multifaceted preference distributions. However, classifier RMs rem...	[ "Jiwoo Hong", "Shao Tang", "Zhipeng Wang" ]	[ "cs.LG", "cs.CL" ]	[ "Computer Science" ]	2026-02-09T00:00:00	https://arxiv.org/abs/2602.08819	https://arxiv.org/pdf/2602.08819v2	2602.08819	10.48550/arXiv.2602.08819	0	0	false	null	arXiv.org	0.4698
45d3d9274615e97b8050cbbb5f3a3ecc738b9f6380f2615e802b5c330e39acd2	[ "arxiv", "semantic_scholar" ]	Learning in Context, Guided by Choice: A Reward-Free Paradigm for Reinforcement Learning with Transformers	In-context reinforcement learning (ICRL) leverages the in-context learning capabilities of transformer models (TMs) to efficiently generalize to unseen sequential decision-making tasks without parameter updates. However, existing ICRL methods rely on explicit reward signals during pretraining, which limits their applic...	[ "Juncheng Dong", "Bowen He", "Moyang Guo", "Ethan X. Fang", "Zhuoran Yang", "Vahid Tarokh" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-02-09T00:00:00	https://arxiv.org/abs/2602.08244	https://arxiv.org/pdf/2602.08244v1	2602.08244	10.48550/arXiv.2602.08244	0	0	false	null	arXiv.org	0.4698
578200056e911dd35d7dc86ec61613a7dc5e8f9d6b2bfc057f1304511c40431c	[ "arxiv", "semantic_scholar" ]	Adversarial Reward Auditing for Active Detection and Mitigation of Reward Hacking	Reinforcement Learning from Human Feedback (RLHF) remains vulnerable to reward hacking, where models exploit spurious correlations in learned reward models to achieve high scores while violating human intent. Existing mitigations rely on static defenses that cannot adapt to novel exploitation strategies. We propose Adv...	[ "Mohammad Beigi", "Ming Jin", "Junshan Zhang", "Qifan Wang", "Lifu Huang" ]	[ "cs.AI", "cs.LG" ]	[ "Computer Science" ]	2026-02-02T00:00:00	https://arxiv.org/abs/2602.01750	https://arxiv.org/pdf/2602.01750v1	2602.01750	10.48550/arXiv.2602.01750	6	0	false	null	arXiv.org	0.4618
54e69b3bd365374b891861a6aae36a8a89cf165afbef9db558a17c9ac4c17840	[ "arxiv", "semantic_scholar" ]	Omni-RRM: Advancing Omni Reward Modeling via Automatic Rubric-Grounded Preference Synthesis	Multimodal large language models (MLLMs) have shown remarkable capabilities, yet their performance is often capped by the coarse nature of existing alignment techniques. A critical bottleneck remains the lack of effective reward models (RMs): existing RMs are predominantly vision-centric, return opaque scalar scores, a...	[ "Zicheng Kong", "Dehua Ma", "Zhenbo Xu", "Alven Yang", "Yiwei Ru", "Haoran Wang", "Zixuan Zhou", "Fuqing Bie", "Liuyu Xiang", "Huijia Wu", "Jian Zhao", "Zhaofeng He" ]	[ "cs.CL" ]	[ "Computer Science" ]	2026-01-31T00:00:00	https://arxiv.org/abs/2602.00846	https://arxiv.org/pdf/2602.00846v1	2602.00846	10.48550/arXiv.2602.00846	6	0	true	null	arXiv.org	0.7101
8dd40a8027df67e156a73e8a83985867627de1e594d465f241e7d247246fe38f	[ "arxiv", "semantic_scholar" ]	Factored Causal Representation Learning for Robust Reward Modeling in RLHF	A reliable reward model is essential for aligning large language models with human preferences through reinforcement learning from human feedback. However, standard reward models are susceptible to spurious features that are not causally related to human labels. This can lead to reward hacking, where high predicted rew...	[ "Yupei Yang", "Lin Yang", "Wanxi Deng", "Lin Qu", "Fan Feng", "Biwei Huang", "Shikui Tu", "Lei Xu" ]	[ "cs.LG" ]	[ "Computer Science" ]	2026-01-29T00:00:00	https://arxiv.org/abs/2601.21350	https://arxiv.org/pdf/2601.21350v2	2601.21350	10.48550/arXiv.2601.21350	1	0	false	null	arXiv.org	0.4572
64de0b1b3f061e1b15a4c4de727cdecdc3a3bb115fa5043446a6e2114765ab6d	[ "arxiv", "semantic_scholar" ]	FunPRM: Function-as-Step Process Reward Model with Meta Reward Correction for Code Generation	Code generation is a core application of large language models (LLMs), yet LLMs still frequently fail on complex programming tasks. Given its success in mathematical reasoning, test-time scaling approaches such as Process Reward Model (PRM)-based Best-of-N selection offer a promising way to improve performance. However...	[ "Ruiyi Zhang", "Peijia Qin", "Qi Cao", "Eric Xue", "Pengtao Xie" ]	[ "cs.LG", "cs.SE" ]	[ "Computer Science" ]	2026-01-29T00:00:00	https://arxiv.org/abs/2601.22249	https://arxiv.org/pdf/2601.22249v1	2601.22249	10.48550/arXiv.2601.22249	3	0	false	null	arXiv.org	0.4572
86b88a864495f1cf6c00ef5d5912c6063e89e3e4ba743ce24e0af6b64c2b66a1	[ "arxiv", "semantic_scholar" ]	The Trajectory Alignment Coefficient in Two Acts: From Reward Tuning to Reward Learning	The success of reinforcement learning (RL) is fundamentally tied to having a reward function that accurately reflects the task objective. Yet, designing reward functions is notoriously time-consuming and prone to misspecification. To address this issue, our first goal is to understand how to support RL practitioners in...	[ "Calarina Muslimani", "Yunshu Du", "Kenta Kawamoto", "Kaushik Subramanian", "Peter Stone", "Peter Wurman" ]	[ "cs.LG", "cs.HC" ]	[ "Computer Science" ]	2026-01-23T00:00:00	https://arxiv.org/abs/2601.16906	https://arxiv.org/pdf/2601.16906v1	2601.16906	10.48550/arXiv.2601.16906	0	0	false	null	arXiv.org	0.4503
177fc0edf41a8ecd68b1a5bf29e9d64dc2a7ef5f9d1d97bbc4279cb23880d5a8	[ "arxiv", "semantic_scholar" ]	Rewarding Creativity: A Human-Aligned Generative Reward Model for Reinforcement Learning in Storytelling	While Large Language Models (LLMs) can generate fluent text, producing high-quality creative stories remains challenging. Reinforcement Learning (RL) offers a promising solution but faces two critical obstacles: designing reliable reward signals for subjective storytelling quality and mitigating training instability. T...	[ "Zhaoyan Li", "Hang Lei", "Yujia Wang", "Lanbo Liu", "Hao Liu", "Liang Yu" ]	[ "cs.AI", "cs.CL" ]	[ "Computer Science" ]	2026-01-12T00:00:00	https://arxiv.org/abs/2601.07149	https://arxiv.org/pdf/2601.07149v1	2601.07149	10.48550/arXiv.2601.07149	1	1	false	null	arXiv.org	0.4377
50619689f08b499da3b28fd371878d0a3c7a5cd8dfe8e0e16bf7aa0333a457a1	[ "arxiv", "semantic_scholar" ]	Reward Modeling from Natural Language Human Feedback	Reinforcement Learning with Verifiable reward (RLVR) on preference data has become the mainstream approach for training Generative Reward Models (GRMs). Typically in pairwise rewarding tasks, GRMs generate reasoning chains ending with critiques and preference labels, and RLVR then relies on the correctness of the prefe...	[ "Zongqi Wang", "Rui Wang", "Yuchuan Wu", "Yiyao Yu", "Pinyi Zhang", "Shaoning Sun", "Yujiu Yang", "Yongbin Li" ]	[ "cs.CL" ]	[ "Computer Science" ]	2026-01-12T00:00:00	https://arxiv.org/abs/2601.07349	https://arxiv.org/pdf/2601.07349v3	2601.07349	10.48550/arXiv.2601.07349	6	1	false	null	arXiv.org	0.4377
1290db498e9302ce8c7d5c4b2740fac1b2288f4e52e1aff0eab040d0138b4c74	[ "arxiv", "semantic_scholar" ]	IRPM: Intergroup Relative Preference Modeling for Pointwise Generative Reward Models	Generative Reward Models (GRMs) have demonstrated strong performance in reward modeling, due to their interpretability and potential for refinement through reinforcement learning (RL). However, widely used pairwise GRMs create a computational bottleneck in reinforcement learning from human feedback (RLHF), when calibra...	[ "Haonan Song", "Qingchen Xie", "Huan Zhu", "Feng Xiao", "Luxi Xing", "Liu Kang", "Fuzhen Li", "Zhiyong Zheng", "Feng Jiang", "Ziheng Li", "Kun Yan", "Qingyi Si", "Yanghua Xiao", "Hongcheng Guo", "Fan Yang" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2026-01-02T00:00:00	https://arxiv.org/abs/2601.00677	https://arxiv.org/pdf/2601.00677v2	2601.00677	10.48550/arXiv.2601.00677	0	0	false	null	arXiv.org	0.4263
9bfa7f674c61a2e0075bfd4810ee32eaf0f8e44d20ffd9eef23b94a23fa5da06	[ "arxiv", "semantic_scholar" ]	Eliminating Inductive Bias in Reward Models with Information-Theoretic Guidance	Reward models (RMs) are essential in reinforcement learning from human feedback (RLHF) to align large language models (LLMs) with human values. However, RM training data is commonly recognized as low-quality, containing inductive biases that can easily lead to overfitting and reward hacking. For example, more detailed ...	[ "Zhuo Li", "Pengyu Cheng", "Zhechao Yu", "Feifei Tong", "Anningzhe Gao", "Tsung-Hui Chang", "Xiang Wan", "Erchao Zhao", "Xiaoxi Jiang", "Guanjun Jiang" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2025-12-29T00:00:00	https://arxiv.org/abs/2512.23461	https://arxiv.org/pdf/2512.23461v2	2512.23461	10.48550/arXiv.2512.23461	6	0	true	https://github.com/Qwen-Applications/DIR	arXiv.org	0.6517
a78990a4fcec3e83dd61b2e41c4845911b97d230532b8bcdd541eb4807047f8e	[ "arxiv", "semantic_scholar" ]	Revisiting the Learning Objectives of Vision-Language Reward Models	Learning generalizable reward functions is a core challenge in embodied intelligence. Recent work leverages contrastive vision language models (VLMs) to obtain dense, domain-agnostic rewards without human supervision. These methods adapt VLMs into reward models through increasingly complex learning objectives, yet mean...	[ "Simon Roy", "Samuel Barbeau", "Giovanni Beltrame", "Christian Desrosiers", "Nicolas Thome" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2025-12-20T00:00:00	https://arxiv.org/abs/2512.20675	https://arxiv.org/pdf/2512.20675v1	2512.20675	10.48550/arXiv.2512.20675	0	0	false	null	arXiv.org	0.4114
6e8eb89a2dd7b852bd82376eddf7ddd1e7652a05dc26b991a1d46d100d8301bb	[ "arxiv", "semantic_scholar" ]	A First-Order Logic-Based Alternative to Reward Models in RLHF	Reinforcement Learning from Human Feedback (RLHF) plays a crucial role in aligning large language models (LLMs) with human values and preferences. However, the quality and stability of the trained reward model largely determine the final alignment performance. Existing approaches such as Proximal Policy Optimization (P...	[ "Chunjin Jian", "Xinhua Zhu" ]	[ "cs.LG", "cs.LO" ]	[ "Computer Science" ]	2025-12-16T00:00:00	https://arxiv.org/abs/2512.14100	https://arxiv.org/pdf/2512.14100v1	2512.14100	10.1109/ICNLP69856.2026.11527781	0	0	true	https://github.com/ChunjinJiang/sgrpo	ICON	0.6286
c9762f1bf7f7c666904cea7812e1215ef6a2b6c83da8894b728a244f92448324	[ "arxiv", "semantic_scholar" ]	Model-Based Reinforcement Learning in Discrete-Action Non-Markovian Reward Decision Processes	Many practical decision-making problems involve tasks whose success depends on the entire system history, rather than on achieving a state with desired properties. Markovian Reinforcement Learning (RL) approaches are not suitable for such tasks, while RL with non-Markovian reward decision processes (NMRDPs) enables age...	[ "Alessandro Trapasso", "Luca Iocchi", "Fabio Patrizi" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2025-12-16T00:00:00	https://arxiv.org/abs/2512.14617	https://arxiv.org/pdf/2512.14617v2	2512.14617	10.48550/arXiv.2512.14617	0	0	false	null	arXiv.org	0.4068
77f31f9101a90c6148d76fbc78f0bf80dbe539c87f8f7181ac1edab5ebf51e20	[ "arxiv", "semantic_scholar" ]	Multi-Objective Reward and Preference Optimization: Theory and Algorithms	This thesis develops theoretical frameworks and algorithms that advance constrained reinforcement learning (RL) across control, preference learning, and alignment of large language models. The first contribution addresses constrained Markov Decision Processes (CMDPs) under the average-cost criterion through the Average...	[ "Akhil Agnihotri" ]	[ "cs.LG" ]	[ "Computer Science" ]	2025-12-11T00:00:00	https://arxiv.org/abs/2512.10601	https://arxiv.org/pdf/2512.10601v1	2512.10601	10.48550/arXiv.2512.10601	0	0	false	null	arXiv.org	0.401
f7b39439ec680bd3ea44113d2831368bb08976abd3ecdd2cc2c38d5f43ff43fd	[ "arxiv", "semantic_scholar" ]	Parent-Guided Semantic Reward Model (PGSRM): Embedding-Based Reward Functions for Reinforcement Learning of Transformer Language Models	We introduce the Parent-Guided Semantic Reward Model (PGSRM), a lightweight reward framework for reinforcement learning (RL) of transformer language models. PGSRM replaces binary correctness signals, human preference data, and trained reward models with a simple signal: cosine similarity between a parent model's refere...	[ "Alexandr Plashchinsky" ]	[ "cs.LG" ]	[ "Computer Science" ]	2025-12-07T00:00:00	https://arxiv.org/abs/2512.06920	https://arxiv.org/pdf/2512.06920v1	2512.06920	10.48550/arXiv.2512.06920	0	0	false	null	arXiv.org	0.3965
6de827cd2b0b9894f239791ed7558a923b8fe3d6aa2ac586df7df0df2f064c7f	[ "arxiv", "semantic_scholar" ]	Average-reward reinforcement learning in semi-Markov decision processes via relative value iteration	This paper applies the authors' recent results on asynchronous stochastic approximation (SA) in the Borkar-Meyn framework to reinforcement learning in average-reward semi-Markov decision processes (SMDPs). We establish the convergence of an asynchronous SA analogue of Schweitzer's classical relative value iteration alg...	[ "Huizhen Yu", "Yi Wan", "Richard S. Sutton" ]	[ "cs.LG", "math.OC" ]	[ "Computer Science", "Mathematics" ]	2025-12-05T00:00:00	https://arxiv.org/abs/2512.06218	https://arxiv.org/pdf/2512.06218v1	2512.06218	10.48550/arXiv.2512.06218	1	0	false	null	arXiv.org	0.3942
24b58c066e5553cd845f8261e95c6ba99a36a3f26d2111eaae2a1221b02b36e1	[ "arxiv", "semantic_scholar" ]	Uncertainty Quantification for Large Language Model Reward Learning under Heterogeneous Human Feedback	We study estimation and statistical inference for reward models used in aligning large language models (LLMs). A key component of LLM alignment is reinforcement learning from human feedback (RLHF), where humans compare pairs of model-generated answers and their preferences are used to train a reward model. However, hum...	[ "Pangpang Liu", "Junwei Lu", "Will Wei Sun" ]	[ "stat.ML", "cs.LG" ]	[ "Computer Science", "Mathematics" ]	2025-12-02T00:00:00	https://arxiv.org/abs/2512.03208	https://arxiv.org/pdf/2512.03208v1	2512.03208	10.48550/arXiv.2512.03208	7	2	false	null	arXiv.org	0.3907
4da791dd8bbc581d9144c98519d29b224f412a983f2ad2b7325fd859fd851971	[ "arxiv", "semantic_scholar" ]	SPARK: Stepwise Process-Aware Rewards for Reference-Free Reinforcement Learning	Process reward models (PRMs) that provide dense, step-level feedback have shown promise for reinforcement learning, yet their adoption remains limited by the need for expensive step-level annotations or ground truth references. We propose SPARK: a three-stage framework where in the first stage a generator model produce...	[ "Salman Rahman", "Sruthi Gorantla", "Arpit Gupta", "Swastik Roy", "Nanyun Peng", "Yang Liu" ]	[ "cs.LG", "cs.AI", "cs.CL" ]	[ "Computer Science" ]	2025-12-02T00:00:00	https://arxiv.org/abs/2512.03244	https://arxiv.org/pdf/2512.03244v1	2512.03244	10.48550/arXiv.2512.03244	0	0	false	null	arXiv.org	0.3907
c476fad52f45bc2b5654fd1cbae91a0178ee36102baab796bd6572298b8de484	[ "arxiv", "semantic_scholar" ]	UMM-RM: An Upcycle-and-Merge MoE Reward Model for Mitigating Reward Hacking	Reward models (RMs) are a critical component of reinforcement learning from human feedback (RLHF). However, conventional dense RMs are susceptible to exploitation by policy models through biases or spurious correlations, resulting in reward hacking: RM scores increase during training while alignment with human preferen...	[ "Lingling Fu", "Yongfu Xue" ]	[ "cs.LG", "cs.IR" ]	[ "Computer Science" ]	2025-11-30T00:00:00	https://arxiv.org/abs/2512.00724	https://arxiv.org/pdf/2512.00724v2	2512.00724	null	0	0	false	null	null	0.2472
7c9dea143b8b749e2c4db110e4996ad16cbad1dfe607511c212853bb407d4c37	[ "arxiv", "semantic_scholar" ]	Reward Engineering for Spatial Epidemic Simulations: A Reinforcement Learning Platform for Individual Behavioral Learning	We present ContagionRL, a Gymnasium-compatible reinforcement learning platform specifically designed for systematic reward engineering in spatial epidemic simulations. Unlike traditional agent-based models that rely on fixed behavioral rules, our platform enables rigorous evaluation of how reward function design affect...	[ "Radman Rakhshandehroo", "Daniel Coombs" ]	[ "cs.LG", "cs.AI", "q-bio.PE" ]	[ "Computer Science", "Biology" ]	2025-11-22T00:00:00	https://arxiv.org/abs/2511.18000	https://arxiv.org/pdf/2511.18000v2	2511.18000	10.48550/arXiv.2511.18000	0	0	true	https://github.com/redradman/ContagionRL	Transactions on Machine Learning Research, 2026	0.5861
0bee6d3826c3610569d98ba75a51489bd84e27e9fbf5cf4ba64475f99818c312	[ "arxiv", "semantic_scholar" ]	The Good, The Bad, and The Hybrid: A Reward Structure Showdown in Reasoning Models Training	Reward design is central to reinforcement learning from human feedback (RLHF) and alignment research. In this work, we propose a unified framework to study hard, continuous, and hybrid reward structures for fine-tuning large language models (LLMs) on mathematical reasoning tasks. Using Qwen3-4B with LoRA fine-tuning on...	[ "Subramanyam Sahoo" ]	[ "cs.LG" ]	[ "Computer Science" ]	2025-11-17T00:00:00	https://arxiv.org/abs/2511.13016	https://arxiv.org/pdf/2511.13016v1	2511.13016	10.48550/arXiv.2511.13016	2	0	false	null	arXiv.org	0.3735
17fb9d747f0085583fe1ed2fa5b050460c94d272e89c861702b1164e991bab9b	[ "arxiv", "semantic_scholar" ]	Probing Preference Representations: A Multi-Dimensional Evaluation and Analysis Method for Reward Models	Previous methods evaluate reward models by testing them on a fixed pairwise ranking test set, but they typically do not provide performance information on each preference dimension. In this work, we address the evaluation challenge of reward models by probing preference representations. To confirm the effectiveness of ...	[ "Chenglong Wang", "Yifu Huo", "Yang Gan", "Yongyu Mu", "Qiaozhi He", "Murun Yang", "Bei Li", "Chunliang Zhang", "Tongran Liu", "Anxiang Ma", "Zhengtao Yu", "Jingbo Zhu", "Tong Xiao" ]	[ "cs.CL" ]	[ "Computer Science" ]	2025-11-16T00:00:00	https://arxiv.org/abs/2511.12464	https://arxiv.org/pdf/2511.12464v1	2511.12464	10.48550/arXiv.2511.12464	1	0	false	null	AAAI Conference on Artificial Intelligence	0.3724
af7324416da21e6c5342ffa6ca1b534996e046adee852042e4ec580a1fbc8374	[ "arxiv", "semantic_scholar" ]	PROF: An LLM-based Reward Code Preference Optimization Framework for Offline Imitation Learning	Offline imitation learning (offline IL) enables training effective policies without requiring explicit reward annotations. Recent approaches attempt to estimate rewards for unlabeled datasets using a small set of expert demonstrations. However, these methods often assume that the similarity between a trajectory and an ...	[ "Shengjie Sun", "Jiafei Lyu", "Runze Liu", "Mengbei Yan", "Bo Liu", "Deheng Ye", "Xiu Li" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2025-11-14T00:00:00	https://arxiv.org/abs/2511.13765	https://arxiv.org/pdf/2511.13765v1	2511.13765	10.48550/arXiv.2511.13765	0	0	false	null	arXiv.org	0.3701
561c5dfc9d4c1ea42581f6dc60d07341cc6a5dca6f5f6ada4ad3fab581832120	[ "arxiv", "semantic_scholar" ]	PIRA: Preference-Oriented Instruction-Tuned Reward Models with Dual Aggregation	Reward models are pivotal for aligning Large Language Models (LLMs) with human preferences. Existing approaches face two key limitations: Discriminative reward models require large-scale annotated data, as they cannot exploit the preference instruction-following capability of LLMs available to generative reward models....	[ "Yongfu Xue" ]	[ "cs.CL" ]	[ "Computer Science" ]	2025-11-14T00:00:00	https://arxiv.org/abs/2511.20668	https://arxiv.org/pdf/2511.20668v2	2511.20668	10.48550/arXiv.2511.20668	1	0	false	null	Conference of the European Chapter of the Association for Computational Linguistics	0.3701
8f189317b9004a615962e67e509af7b4cde256222f8c864355e37070f2533d8d	[ "arxiv", "semantic_scholar" ]	Debiasing Reward Models by Representation Learning with Guarantees	Recent alignment techniques, such as reinforcement learning from human feedback, have been widely adopted to align large language models with human preferences by learning and leveraging reward models. In practice, these models often exploit spurious correlations, involving, e.g., response length, discrimination, sycop...	[ "Ignavier Ng", "Patrick Blöbaum", "Siddharth Bhandari", "Kun Zhang", "Shiva Kasiviswanathan" ]	[ "cs.LG", "cs.AI", "stat.ML" ]	[ "Computer Science", "Mathematics" ]	2025-10-27T00:00:00	https://arxiv.org/abs/2510.23751	https://arxiv.org/pdf/2510.23751v1	2510.23751	10.48550/arXiv.2510.23751	3	0	false	null	arXiv.org	0.3495
4c9fe65e0624d37e8f46f3e953777ae447e302a5e1206d86df0e5a1eb4f51104	[ "arxiv", "semantic_scholar" ]	Rectifying Shortcut Behaviors in Preference-based Reward Learning	In reinforcement learning from human feedback, preference-based reward models play a central role in aligning large language models to human-aligned behavior. However, recent studies show that these models are prone to reward hacking and often fail to generalize well due to over-optimization. They achieve high reward s...	[ "Wenqian Ye", "Guangtao Zheng", "Aidong Zhang" ]	[ "cs.AI", "cs.LG" ]	[ "Computer Science" ]	2025-10-21T00:00:00	https://arxiv.org/abs/2510.19050	https://arxiv.org/pdf/2510.19050v1	2510.19050	10.48550/arXiv.2510.19050	4	0	false	null	arXiv.org	0.3426
b6934406d685908450255a2122583c3770a56586d7640b93eacb6670c89bd5c0	[ "arxiv", "semantic_scholar" ]	Auto-Rubric: Learning From Implicit Weights to Explicit Rubrics for Reward Modeling	Conventional reward modeling relies on gradient descent over neural weights, creating opaque, data-hungry "black boxes." We propose a paradigm shift from implicit to explicit reward parameterization, recasting optimization from continuous weight spaces to the discrete space of natural language rubrics. We introduce a t...	[ "Lipeng Xie", "Sen Huang", "Zhuo Zhang", "Anni Zou", "Yunpeng Zhai", "Dingchao Ren", "Kezun Zhang", "Haoyuan Hu", "Boyin Liu", "Haoran Chen", "Zhaoyang Liu", "Bolin Ding" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2025-10-20T00:00:00	https://arxiv.org/abs/2510.17314	https://arxiv.org/pdf/2510.17314v2	2510.17314	null	16	1	false	null	null	0.3076
96f08c4422453cf7c153b6fe82cbe99deb9eb5cfe747c801d30a088826b59fa4	[ "arxiv", "semantic_scholar" ]	Learning Correlated Reward Models: Statistical Barriers and Opportunities	Random Utility Models (RUMs) are a classical framework for modeling user preferences and play a key role in reward modeling for Reinforcement Learning from Human Feedback (RLHF). However, a crucial shortcoming of many of these techniques is the Independence of Irrelevant Alternatives (IIA) assumption, which collapses \...	[ "Yeshwanth Cherapanamjeri", "Constantinos Daskalakis", "Gabriele Farina", "Sobhan Mohammadpour" ]	[ "cs.LG", "econ.EM", "stat.ML" ]	[ "Computer Science", "Economics", "Mathematics" ]	2025-10-17T00:00:00	https://arxiv.org/abs/2510.15839	https://arxiv.org/pdf/2510.15839v2	2510.15839	10.48550/arXiv.2510.15839	1	0	false	null	arXiv.org	0.338
00de8e8d4cf4b485f144a656c107a4e206080b2875bf5ecd5588be0fa939d356	[ "arxiv", "semantic_scholar" ]	Reinforcement Learning with Stochastic Reward Machines	Reward machines are an established tool for dealing with reinforcement learning problems in which rewards are sparse and depend on complex sequences of actions. However, existing algorithms for learning reward machines assume an overly idealized setting where rewards have to be free of noise. To overcome this practical...	[ "Jan Corazza", "Ivan Gavran", "Daniel Neider" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2025-10-16T00:00:00	https://arxiv.org/abs/2510.14837	https://arxiv.org/pdf/2510.14837v1	2510.14837	10.1609/aaai.v36i6.20594	37	1	true	https://github.com/corazza/srm	AAAI Conference on Artificial Intelligence	0.5206
e56b1b6cbaaea933438182ff059046b34c2f406f22e68cf5fad1dc637af1f7aa	[ "arxiv", "semantic_scholar" ]	Information-Theoretic Reward Modeling for Stable RLHF: Detecting and Mitigating Reward Hacking	Despite the success of Reinforcement Learning from Human Feedback (RLHF) in aligning language models with human values, reward hacking-or reward over-optimization-remains a major challenge. We identify two key obstacles to its mitigation: (1) reward misgeneralization in reward modeling, where reward models overfit to s...	[ "Yuchun Miao", "Liang Ding", "Sen Zhang", "Rong Bao", "Lefei Zhang", "Dacheng Tao" ]	[ "cs.LG" ]	[ "Computer Science" ]	2025-10-15T00:00:00	https://arxiv.org/abs/2510.13694	https://arxiv.org/pdf/2510.13694v1	2510.13694	10.48550/arXiv.2510.13694	5	0	false	null	arXiv.org	0.3357
64ea1df719f224643cf9e9195b5105dfc1317abcd4d662f99fb0e3f92909daf8	[ "arxiv", "semantic_scholar" ]	Repairing Reward Functions with Feedback to Mitigate Reward Hacking	Human-designed reward functions for reinforcement learning (RL) agents are frequently misaligned with the humans' true, unobservable objectives, and thus act only as proxies. Optimizing for a misspecified proxy reward function often induces reward hacking, resulting in a policy misaligned with the human's true objectiv...	[ "Stephane Hatgis-Kessell", "Logan Mondal Bhamidipaty", "Emma Brunskill" ]	[ "cs.AI", "cs.LG" ]	[ "Computer Science" ]	2025-10-14T00:00:00	https://arxiv.org/abs/2510.13036	https://arxiv.org/pdf/2510.13036v2	2510.13036	null	0	0	false	null	null	0.2129
fbba9c49fbdca28416d16c7bf8e57d8d347c1cf9d77608543082a1f869ddb0f7	[ "arxiv", "semantic_scholar" ]	APLOT: Robust Reward Modeling via Adaptive Preference Learning with Optimal Transport	The reward model (RM) plays a crucial role in aligning Large Language Models (LLMs) with human preferences through Reinforcement Learning, where the Bradley-Terry (BT) objective has been recognized as simple yet powerful, specifically for pairwise preference learning. However, BT-based RMs often struggle to effectively...	[ "Zhuo Li", "Yuege Feng", "Dandan Guo", "Jinpeng Hu", "Anningzhe Gao", "Xiang Wan" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2025-10-13T00:00:00	https://arxiv.org/abs/2510.10963	https://arxiv.org/pdf/2510.10963v1	2510.10963	10.18653/v1/2025.emnlp-main.281	5	0	true	https://github.com/BIRlz/APLOT	Conference on Empirical Methods in Natural Language Processing	0.5153
d6fc5e2cdeba9ad1434fe8b175796eaee28a8cb82177135a4928bebb698b19e5	[ "arxiv", "semantic_scholar" ]	Reward Model Perspectives: Whose Opinions Do Reward Models Reward?	Reward models (RMs) are central to the alignment of language models (LMs). An RM often serves as a proxy for human preferences to guide downstream LM behavior. However, our understanding of RM behavior is limited. Our work (i) formalizes a framework for measuring the alignment of opinions captured by RMs, (ii) investig...	[ " Elle" ]	[ "cs.CL", "cs.AI" ]	[ "Computer Science" ]	2025-10-07T00:00:00	https://arxiv.org/abs/2510.06391	https://arxiv.org/pdf/2510.06391v1	2510.06391	10.18653/v1/2025.emnlp-main.754	1	0	false	null	Conference on Empirical Methods in Natural Language Processing	0.3266
2742adcbf6ac3f20d94253aa313855099aef63d713d48f1f07bf358570660e32	[ "arxiv", "semantic_scholar" ]	Margin Adaptive DPO: Leveraging Reward Model for Granular Control in Preference Optimization	Direct Preference Optimization (DPO) has emerged as a simple and effective method for aligning large language models. However, its reliance on a fixed temperature parameter leads to suboptimal training on diverse preference data, causing overfitting on easy examples and under-learning from informative ones. Recent meth...	[ "Hyung Gyu Rho" ]	[ "cs.LG", "cs.AI" ]	[ "Computer Science" ]	2025-10-06T00:00:00	https://arxiv.org/abs/2510.05342	https://arxiv.org/pdf/2510.05342v2	2510.05342	10.48550/arXiv.2510.05342	1	0	false	null	arXiv.org	0.3254
44421c4efbd4afb81532a01e87ac77639c15c18f1a70922f1444ac43d3502ae5	[ "arxiv", "semantic_scholar" ]	Limited Preference Data? Learning Better Reward Model with Latent Space Synthesis	Reward modeling, crucial for aligning large language models (LLMs) with human preferences, is often bottlenecked by the high cost of preference data. Existing textual data synthesis methods are computationally expensive. We propose a novel framework LENS for synthesizing preference data directly in the LLM's latent emb...	[ "Leitian Tao", "Xuefeng Du", "Sharon Li" ]	[ "cs.CL" ]	[ "Computer Science" ]	2025-09-30T00:00:00	https://arxiv.org/abs/2509.26074	https://arxiv.org/pdf/2509.26074v2	2509.26074	10.48550/arXiv.2509.26074	2	0	true	https://github.com/deeplearning-wisc/lens	arXiv.org	0.4923

End of preview. Expand in Data Studio

Reward Modeling Papers — FineSet

A research-paper dataset on Reward Modeling Papers, assembled, deduplicated, and quality-scored by FineSet from arXiv and Semantic Scholar.

📸 This is a dated snapshot — generated 2026-06-19. It is not auto-updated. Research on Reward Modeling Papers moves fast — new papers land on arXiv every week. Want this same dataset refreshed daily, on a topic you choose? See the bottom. ↓

Why this dataset

Quality-scored: quality_score float (0–1), blends citations with recency + code/venue signals — filter out the noise
Papers with code: 86 flagged via has_code — find reproducible work fast
Deduplicated: arXiv + Semantic Scholar cross-referenced, duplicate records merged
Clean JSONL: 460 records, one per line, normalized fields — no encoding garbage

Dataset details

Records: 460
Date range: 2020–2026
Snapshot date: 2026-06-19 (frozen — see note above)
Sources: arXiv, Semantic Scholar (cross-referenced, duplicates merged)
arXiv categories: cs.LG, cs.CL
Quality scoring: citations + recency + code/venue blend, 0–1 (p50=0.325, p90=0.588)
Format: JSONL, one record per line

Fields

Field	Type	Description
id	string	Deterministic SHA256 record id
sources	list	Which sources contributed (`arxiv`, `semantic_scholar`)
title	string	Paper title
abstract	string	Full abstract
authors	list	Author names
categories	list	arXiv category codes
fields_of_study	list	Semantic Scholar field tags
published_date	string	ISO 8601 date
url	string	arXiv abstract URL
pdf_url	string\|null	Open-access PDF if available
arxiv_id	string\|null	arXiv identifier
doi	string\|null	DOI if available
citation_count	int	Citation count (Semantic Scholar)
influential_citation_count	int	Influential citations (Semantic Scholar)
has_code	bool	Code repo detected in the arXiv comment
code_url	string\|null	GitHub URL if detected
venue	string\|null	Publication venue
quality_score	float	0–1, blended (citations + recency + code/venue)

Quality score methodology

quality_score = max(impact, freshness), clamped to [0, 1], where:

impact = max( log10(citations+1)/4 , log10(influential_citations+1)/2 ) — realized impact (0.5 at 100 citations, ~0.75 at 1,000, 1.0 at 10,000+).
freshness = recency × (0.35 + 0.30·has_code + 0.20·has_venue) — a baseline for recent papers (so a strong paper published this week isn't scored 0 just for lacking citations), where recency is 1.0 for papers ≤60 days old and decays linearly to 0 by ~18 months.

Old highly-cited papers score on impact; brand-new papers score on freshness; old uncited papers score ~0. Useful for filtering training data by quality, not just age.

👉 Want this on YOUR topic, updated daily?

This snapshot is frozen at 2026-06-19. The live FineSet pipeline keeps a dataset like this refreshed every day on whatever topic you describe — new papers in, dedup and quality scoring automatic, export as JSONL/Parquet or push straight to the Hub.

Tell me the topic you'd want and I'll run the pipeline on it — open a discussion on this dataset, it's free and it's how I decide what to build next.

→ fineset.io — describe what you want to train on, get a dataset. Early-access waitlist open (referral skip available).

Downloads last month: 29