语言大模型 / LLM - arXivDaily 专题

2303.02468 2024-01-05 cs.CL cs.LG 62%

Lon-ea at SemEval-2023 Task 11: A Comparison of Activation Functions for Soft and Hard Label Prediction

Peyman Hosseini, Mehran Hosseini, Sana Sabah Al-Azzawi, Marcus Liwicki, Ignacio Castro, Matthew Purver

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.CL、cs.LG

Comments Accepted in ACL 2023 SemEval Workshop as selected task paper

2308.06385 2023-12-15 cs.CL cs.AI 62%

ZYN: Zero-Shot Reward Models with Yes-No Questions for RLAIF

Victor Gallego

专题命中后训练与偏好优化：language model(abstract)；分类 cs.CL、cs.AI

Comments pre-print, work in progress

2310.11518 2023-11-30 cs.GT cs.AI cs.LG 62%

Guarantees for Self-Play in Multiplayer Games via Polymatrix Decomposability

Revan MacQueen, James R. Wright

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.AI、cs.LG

Comments To appear at NeurIPS 2023

2310.12036 2023-11-23 cs.AI cs.LG stat.ML 62%

A General Theoretical Paradigm to Understand Learning from Human Preferences

Mohammad Gheshlaghi Azar, Mark Rowland, Bilal Piot, Daniel Guo, Daniele Calandriello, Michal Valko, Rémi Munos

专题命中后训练与偏好优化：RLHF(abstract)；分类 cs.AI、cs.LG

2311.04919 2023-11-10 cs.CL cs.AI cs.HC 62%

The Impact of Preference Agreement in Reinforcement Learning from Human Feedback: A Case Study in Summarization

Sian Gooding, Hassan Mansoor

专题命中后训练与偏好优化：RLHF(abstract)；分类 cs.CL、cs.AI

2310.20077 2023-11-01 cs.CL cs.LG 62%

Partial Tensorized Transformers for Natural Language Processing

Subhadra Vadlamannati, Ryan Solgi

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.CL、cs.LG

Comments In Review under the 16th International Conference on Agents and Artificial Intelligence

2206.02231 2023-09-08 cs.LG cs.AI cs.SY eess.SY 62%

Models of human preference for learning reward functions

W. Bradley Knox, Stephane Hatgis-Kessell, Serena Booth, Scott Niekum, Peter Stone, Alessandro Allievi

专题命中后训练与偏好优化：RLHF(abstract)；分类 cs.AI、cs.LG

Comments 16 pages (40 pages with references and appendix), 23 figures

2309.00709 2023-09-06 cs.AI cs.LG cs.RO 62%

Reinforcement Learning with Human Feedback for Realistic Traffic Simulation

Yulong Cao, Boris Ivanovic, Chaowei Xiao, Marco Pavone

专题命中后训练与偏好优化：RLHF(abstract)；分类 cs.AI、cs.LG

Comments 9 pages, 4 figures

2308.07452 2023-08-16 cs.LG cs.AI 62%

GRU-D-Weibull: A Novel Real-Time Individualized Endpoint Prediction

Xiaoyang Ruan, Liwei Wang, Charat Thongprayoon, Wisit Cheungpasitporn, Hongfang Liu

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.AI、cs.LG

Comments 30 pages, 7 figures, 4 supplementary figures

2305.18438 2023-07-04 cs.LG cs.AI math.OC math.ST stat.ML stat.TH 62%

Reinforcement Learning with Human Feedback: Learning Dynamic Choices via Pessimism

Zihao Li, Zhuoran Yang, Mengdi Wang

专题命中后训练与偏好优化：RLHF(abstract)；分类 cs.AI、cs.LG

2305.07036 2023-05-15 cs.LG cs.AI 62%

GFlowNets with Human Feedback

Yinchuan Li, Shuang Luo, Yunfeng Shao, Jianye Hao

专题命中后训练与偏好优化：RLHF(abstract)；分类 cs.AI、cs.LG

2303.02891 2023-03-07 cs.CY cs.AI cs.LG 62%

Perspectives on the Social Impacts of Reinforcement Learning with Human Feedback

Gabrielle Kaili-May Liu

专题命中后训练与偏好优化：RLHF(abstract)；分类 cs.AI、cs.LG

2210.15906 2023-03-01 cs.AI cs.HC cs.LG 62%

Relative Behavioral Attributes: Filling the Gap between Symbolic Goal Specification and Reward Learning from Human Preferences

Lin Guan, Karthik Valmeekam, Subbarao Kambhampati

专题命中后训练与偏好优化：RLHF(abstract)；分类 cs.AI、cs.LG

Comments ICLR 2023 Camera Ready

2209.14375 2022-09-30 cs.LG cs.CL 62%

Improving alignment of dialogue agents via targeted human judgements

Amelia Glaese, Nat McAleese, Maja Trębacz, John Aslanides, Vlad Firoiu, Timo Ewalds, Maribeth Rauh, Laura Weidinger, Martin Chadwick, Phoebe Thacker, Lucy Campbell-Gillingham, Jonathan Uesato, Po-Sen Huang, Ramona Comanescu, Fan Yang, Abigail See, Sumanth Dathathri, Rory Greig, Charlie Chen, Doug Fritz, Jaume Sanchez Elias, Richard Green, Soňa Mokrá, Nicholas Fernando, Boxi Wu, Rachel Foley, Susannah Young, Iason Gabriel, William Isaac, John Mellor, Demis Hassabis, Koray Kavukcuoglu, Lisa Anne Hendricks, Geoffrey Irving

专题命中后训练与偏好优化：language model(abstract)；分类 cs.CL、cs.LG

2203.01156 2022-07-12 cs.LG cs.AI 62%

Engineering the Neural Automatic Passenger Counter

Nico Jahn, Michael Siebert

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.AI、cs.LG

Comments 12 pages, 8 figures

Journal ref j.engappai.2022.105148

1902.01996 2022-03-10 stat.ML cs.AI cs.LG 62%

Are All Layers Created Equal?

Chiyuan Zhang, Samy Bengio, Yoram Singer

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.AI、cs.LG

Comments JMLR 2022, 28 pages, 21 figures

2108.00106 2021-12-08 cs.LG cs.AI 62%

Soft Calibration Objectives for Neural Networks

Archit Karandikar, Nicholas Cain, Dustin Tran, Balaji Lakshminarayanan, Jonathon Shlens, Michael C. Mozer, Becca Roelofs

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.AI、cs.LG

Comments 17 pages total, 10 page main paper, 5 page appendix, 10 figures total, 8 figures in main paper, 2 figures in appendix

2110.02550 2021-10-26 cs.LG cs.AI 62%

CBP: Backpropagation with constraint on weight precision using a pseudo-Lagrange multiplier method

Guhyun Kim, Doo Seok Jeong

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.AI、cs.LG

Comments Accepted. NeurIPS 2021. The code is available at https://github.com/dooseokjeong/CBP

2109.04727 2021-09-13 cs.CL cs.AI 62%

A Simple and Effective Method To Eliminate the Self Language Bias in Multilingual Representations

Ziyi Yang, Yinfei Yang, Daniel Cer, Eric Darve

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.CL、cs.AI

Comments Accepted to the 2021 Conference on Empirical Methods in Natural Language Processing

2004.13796 2021-04-28 cs.CL cs.LG 62%

TextGAIL: Generative Adversarial Imitation Learning for Text Generation

Qingyang Wu, Lei Li, Zhou Yu

专题命中后训练与偏好优化：language model(abstract)；分类 cs.CL、cs.LG

Comments AAAI 2021

2103.06506 2021-03-18 cs.ET cs.AI cs.AR cs.LG 62%

Memristive Stochastic Computing for Deep Learning Parameter Optimization

Corey Lammie, Jason K. Eshraghian, Wei D. Lu, Mostafa Rahimi Azghadi

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.AI、cs.LG

Comments Accepted by IEEE Transactions on Circuits and Systems Part II: Express Briefs

Journal ref IEEE Transactions on Circuits and Systems Part II: Express Briefs, 2021

1904.06312 2019-04-15 cs.LG cs.AI stat.ML 62%

Let's Play Again: Variability of Deep Reinforcement Learning Agents in Atari Environments

Kaleigh Clary, Emma Tosch, John Foley, David Jensen

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.AI、cs.LG

Comments NeurIPS 2018 Critiquing and Correcting Trends Workshop

2402.07314 2024-11-13 cs.LG stat.ML 61%

Online Iterative Reinforcement Learning from Human Feedback with General Preference Model

Chenlu Ye, Wei Xiong, Yuheng Zhang, Hanze Dong, Nan Jiang, Tong Zhang

专题命中后训练与偏好优化：RLHF(abstract,comments)；分类 cs.LG

Comments RLHF, Preference Learning, Alignment for LLMs

2310.17303 2024-06-11 stat.ML cs.LG 61%

Demonstration-Regularized RL

Daniil Tiapkin, Denis Belomestny, Daniele Calandriello, Eric Moulines, Alexey Naumov, Pierre Perrault, Michal Valko, Pierre Menard

专题命中后训练与偏好优化：RLHF(abstract,comments)；分类 cs.LG

Comments This revision fixes an error due to use of some incorrect results (Lemma 32, Corollary 11 by Talebi & Maillard, 2018) in the proof of Theorem 8. The condition for the RLHF results have slightly changed

2608.20913 2026-08-24 cs.CV cs.AI 新提交 57%

Explainable Deepfake Detection with Feature-robust Augmentation and Evidence-grounded Explanation Optimization

基于特征鲁棒增强与证据支撑的解释性深度伪造检测

Zhu Xu, Jiaqi Tang, Pokai Chen, Yuxin Peng, Yang Liu

机构 * Wangxuan Institute of Computer Technology, Peking University（北京大学王选计算机研究所）

专题命中后训练与偏好优化：preference optimization(abstract)；分类 cs.AI

AI总结该研究针对深度伪造检测的鲁棒性与可解释性缺陷，提出含特征鲁棒增强、证据支撑偏好优化的框架，在ACM Multimedia 2026相关任务中获第一名。

2608.19540 2026-08-21 cs.LG cs.CV 新提交 57%

Continuous Adversarial MeanFlow Transfer

连续对抗平均流迁移

Yara Bahram, Zahra Dehghani, Mélodie Desbos, Eric Granger, Pablo Piantanida, Mohammadhadi Shateri

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.LG

AI总结本研究提出MeanFlow-Transfer与Continuous Adversarial MeanFlow，解决有限数据下新域生成器训练的适配与加速问题，在四个源模型适配五个目标域时，FID等指标相当或更优且NFEs最多减125倍，少步FID平均提29%。

Comments Paper under review

2608.18237 2026-08-20 stat.ML cs.LG math.OC 新提交 57%

Sobolev Regularized Score Difference Estimation in Diffusion Models

扩散模型中的Sobolev正则化得分差估计

Chenghan Xie, Jose Blanchet, Renyuan Xu

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.LG

AI总结本文针对扩散模型得分差估计存在的统计一致性或可扩展性问题，提出Sobolev正则化的一致可扩展估计器，在小样本场景稳定性提升，真实任务性能优于非正则化方法。

Comments Accpeted by ICML 2026

2603.24126 2026-08-19 cs.LG cs.PL 版本更新 57%

Likelihood Hacking in Probabilistic Program Synthesis

概率程序合成中的似然黑客行为

Jacek Karwowski, Younesse Kaddar, Zihuiwen Ye, Esmeralda S. Whitammer, Sam Staton

机构 * University of Oxford, Department of Computer Science（牛津大学计算机科学系）； University of Edinburgh, School of Informatics（爱丁堡大学信息学院）； CIFAR Fellow, Learning in Machines and Brains（CIFAR Fellow, 机器与大脑学习）

专题命中后训练与偏好优化：language model(abstract)；分类 cs.LG

AI总结研究概率程序合成中语言模型通过强化学习生成程序时可能人为夸大边际似然奖励的问题，提出安全语言片段SafeStan以防止似然黑客行为。

Journal ref Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence, PMLR 337:2744-2791, 2026

2608.16473 2026-08-18 cs.LG cs.NA math.NA 新提交 57%

Reference-free logged energy-oracle recovery for neural approximations of symmetric coercive variational problems: conforming Riesz reconstruction and archive-level selection

对称强制变分问题神经近似的无参考记录能量-神谕恢复：符合里泽重构与存档级选择

Karim Bounja, Lahcen Laayouni, Boujemaa Achchab, Abdeljalil Sakat

机构 * Laboratory for Analysis and Modeling of Systems and Decision Support (LAMSAD), Hassan 1st University of Settat（塞塔特哈桑一世大学系统与决策支持分析建模实验室）； Al Akhawayn University in Ifrane（伊夫兰阿赫拉万大学）

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.LG

AI总结针对对称强制变分问题的神经近似，提出无参考的记录能量神谕恢复准则，通过符合里泽监测器实现存档级选择，在扩散、弹性等问题上验证了其有效性。

Comments 35 pages, 7 figures

2605.26182 2026-08-18 cs.AI cs.GR 版本更新 57%

BrickAnything: Geometry-Conditioned Buildable Brick Generation with Structure-Aware Tokenization

BrickAnything: 基于几何条件的可构建砖块生成与结构感知标记化

Zhengyang Ni, Feng Yan, Yu Guo, Fei Wang

机构 * Xi’an Jiaotong University（西安交通大学）； State Key Laboratory of Human-Machine Hybrid Augmented Intelligence（人机混合增强智能国家重点实验室）； Institute of Artificial Intelligence and Robotics（人工智能与机器人研究院）

专题命中后训练与偏好优化：post-training(abstract)；分类 cs.AI

AI总结提出BrickAnything，一个基于几何条件的自回归框架，通过结构感知树标记化生成满足装配约束和结构稳定性的砖块结构。

Comments Revised version with updated Code: https://github.com/xjtunzy/BrickAnything

1. 后训练与偏好优化 4541 篇

Lon-ea at SemEval-2023 Task 11: A Comparison of Activation Functions for Soft and Hard Label Prediction

ZYN: Zero-Shot Reward Models with Yes-No Questions for RLAIF

Guarantees for Self-Play in Multiplayer Games via Polymatrix Decomposability

A General Theoretical Paradigm to Understand Learning from Human Preferences

The Impact of Preference Agreement in Reinforcement Learning from Human Feedback: A Case Study in Summarization

Partial Tensorized Transformers for Natural Language Processing

Models of human preference for learning reward functions

Reinforcement Learning with Human Feedback for Realistic Traffic Simulation

GRU-D-Weibull: A Novel Real-Time Individualized Endpoint Prediction

Reinforcement Learning with Human Feedback: Learning Dynamic Choices via Pessimism

GFlowNets with Human Feedback

Perspectives on the Social Impacts of Reinforcement Learning with Human Feedback

Relative Behavioral Attributes: Filling the Gap between Symbolic Goal Specification and Reward Learning from Human Preferences

Improving alignment of dialogue agents via targeted human judgements

Engineering the Neural Automatic Passenger Counter

Are All Layers Created Equal?

Soft Calibration Objectives for Neural Networks

CBP: Backpropagation with constraint on weight precision using a pseudo-Lagrange multiplier method

A Simple and Effective Method To Eliminate the Self Language Bias in Multilingual Representations

TextGAIL: Generative Adversarial Imitation Learning for Text Generation

Memristive Stochastic Computing for Deep Learning Parameter Optimization

Let's Play Again: Variability of Deep Reinforcement Learning Agents in Atari Environments

Online Iterative Reinforcement Learning from Human Feedback with General Preference Model

Demonstration-Regularized RL

Explainable Deepfake Detection with Feature-robust Augmentation and Evidence-grounded Explanation Optimization

Continuous Adversarial MeanFlow Transfer

Sobolev Regularized Score Difference Estimation in Diffusion Models

Likelihood Hacking in Probabilistic Program Synthesis

Reference-free logged energy-oracle recovery for neural approximations of symmetric coercive variational problems: conforming Riesz reconstruction and archive-level selection

BrickAnything: Geometry-Conditioned Buildable Brick Generation with Structure-Aware Tokenization