Peiyang Liu 刘培阳刘培阳 Peiyang Liu

Ph.D. Candidate 博士在读

Peking University · School of Software and Microelectronics · National Engineering Research Center for Software Engineering 北京大学 · 软件与微电子学院 · 软件工程国家工程研究中心

Building Trustworthy RAG — retrieval-augmented systems that are faithful, robust, attributable, secure, and efficient. 致力于构建可信赖 RAG（Trustworthy RAG） ——忠实、鲁棒、可归因、安全、高效的检索增强生成系统。

258

Citations

Publications

CCF-A

CCF-B

About Me关于我

Hi! I'm Peiyang Liu, a first-year Ph.D. student at the School of Software and Microelectronics, Peking University, advised by Prof. Wei Ye in The Knowledge Computing Lab, National Engineering Research Center for Software Engineering.

My research mission is to build Trustworthy RAG (T-RAG) — retrieval-augmented systems we can actually deploy in high-stakes domains (medicine, law, finance). Around this mission, I work along five pillars: Faithfulness, Robustness, Attributability, Security, and Efficiency. I am also actively working on LLM reasoning & post-training and large-scale information retrieval.

你好！我是刘培阳，目前博士一年级，就读于北京大学软件与微电子学院，导师是软件工程国家工程研究中心知识计算实验室的叶蔚老师。

我的研究使命是构建 可信赖 RAG（Trustworthy RAG, T-RAG）——让检索增强生成系统真正能落地于医疗、法律、金融等高风险场景。围绕这一使命，我从五个支柱展开工作：忠实性、鲁棒性、可归因性、安全性与高效性。同时也在大模型推理与后训练与大规模信息检索方向持续探索。

Research研究方向

FLAGSHIP旗舰方向

Trustworthy RAG (T-RAG) 可信赖 RAG (T-RAG)

I build retrieval-augmented generation systems along five pillars so they can be safely deployed in high-stakes domains — medicine, law, and finance. 我以五个支柱构建可信赖的 RAG 系统，使其能够部署于医疗、法律、金融等高风险场景。

🎯

Faithfulness

忠实性

Answers stay faithful to retrieved evidence, no hallucination. 答案忠实于检索证据，不产生幻觉。

🛡️

Robustness

鲁棒性

Resilient to noisy, counterfactual, or adversarial retrievals. 面对噪声、反事实、对抗检索结果仍能稳定。

🔍

Attributability

可归因性

Every claim traces back to pixel- or token-level evidence. 每一条论据可追溯至像素或 token 级证据。

🔒

Security

安全性

Defends against unauthorized data theft and knowledge-base poisoning. 防御未授权数据窃取与知识库投毒。

⚡

Efficiency

高效性

Trustworthy yet deployable — compressed context, scalable inference. 可信赖与可部署并存——上下文压缩、可扩展推理。

Faithful RAG Robust RAG Attributable RAG Secure RAG Multimodal RAG Context Compression

LLM Reasoning & Post-Training 大模型推理与后训练

Improving how LLMs learn and reason after pretraining — synthesizing reasoning paths from search trajectories, diagnosing SFT failure modes, designing efficient reward models, and calibrating code generation. 提升大模型在预训练之后的学习与推理能力——从搜索轨迹合成推理路径、诊断 SFT 失效模式、设计高效奖励模型，到代码生成的层级校准。

Reasoning Reward Model SFT Analysis Code Generation Long Context

Information Retrieval & Text Embedding 信息检索与文本嵌入

Large-scale dense retrieval with embedding optimization — label enhancement & smoothing, knowledge distillation, data quality control, and cross-modal retrieval. 面向大规模稠密检索的嵌入优化——标签增强与平滑、知识蒸馏、数据质量控制以及跨模态检索。

Dense Retrieval Text Embedding Label Quality Knowledge Distillation Video Search

Publications发表论文

* denotes equal contribution · blue highlight denotes myself * 表示共同一作 · 蓝色高亮为本人

No publications match the selected filters. 没有符合当前筛选条件的论文。

2026

Learning from Contrasts: Synthesizing Reasoning Paths from Diverse Search Trajectories

Peiyang Liu, Zhirui Chen, Xi Wang, Di Liang, Youru Li, Zhi Cai, Wei Ye

ACL 2026CCF-AOral The 64th Annual Meeting of the Association for Computational Linguistics

arXiv

NeuroSym-Cal: Bridging the Reasoning-Execution Gap in Code Generation via Hierarchical Calibration

Peiyang Liu, Yining Wang, Youru Li, Long Li, Zhi Cai, Wei Ye

ACL 2026Findings Findings of the Association for Computational Linguistics: ACL 2026

StructKV: Preserving the Structural Skeleton for Scalable Long-Context Inference

Zhirui Chen, Peiyang Liu, Ling Shao

ACL 2026Findings Findings of the Association for Computational Linguistics: ACL 2026

arXiv

Why Supervised Fine-Tuning Fails to Learn: A Systematic Study of Incomplete Learning in Large Language Models

Chao Xue, Yao Wang, Mengqiao Liu, Di Liang, Xingsheng Han, Peiyang Liu, Xianjie Wu, Chenyao Lu, Lei Jiang, Yu Lu, Haibo Shi, Shuang Liang, Minlong Peng, Flora D. Salim

ACL 2026CCF-A The 64th Annual Meeting of the Association for Computational Linguistics

arXiv

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

Chao Xue, Yao Wang, Mengqiao Liu, Di Liang, Xingsheng Han, Peiyang Liu, Xianjie Wu, Chenyao Lu, Lei Jiang, Yu Lu, Haibo Shi, Shuang Liang, Minlong Peng, Flora D. Salim

ACL 2026Findings Findings of the Association for Computational Linguistics: ACL 2026

arXiv

Parameter Importance is Not Static: Evolving Parameter Isolation for Supervised Fine-Tuning

Zekai Lin, Chao Xue, Di Liang, Xingsheng Han, Peiyang Liu, Xianjie Wu, Lei Jiang, Yu Lu, Bob Simons, Shuang Liang, Minlong Peng

ACL 2026CCF-AOral The 64th Annual Meeting of the Association for Computational Linguistics

arXiv

Chain of Evidence: Pixel-Level Visual Attribution for Iterative Retrieval-Augmented Generation

Peiyang Liu, Ziqiang Cui, Xi Wang, Di Liang, Wei Ye

SIGIR 2026CCF-AOral Proceedings of the 49th International ACM SIGIR Conference on Research and Development in Information Retrieval

arXiv

Beyond Semantic Relevance: Counterfactual Risk Minimization for Robust Retrieval-Augmented Generation

Peiyang Liu, Qiang Yan, Ziqiang Cui, Di Liang, Xi Wang, Wei Ye

SIGIR 2026CCF-AOral Proceedings of the 49th International ACM SIGIR Conference on Research and Development in Information Retrieval

arXiv

ToolSafe: Enhancing Tool Invocation Safety of LLM-based Agents via Proactive Step-level Guardrail and Feedback

Yutao Mou, Zhangchi Xue, Lijun Li, Peiyang Liu, Shikun Zhang, Wei Ye, Jing Shao

ACL 2026Findings Findings of the Association for Computational Linguistics: ACL 2026

arXiv

SQL-ASTRA: Alleviating Sparse Feedback in Agentic SQL via Column-Set Matching and Trajectory Aggregation

Long Li, Zhijian Zhou, Jiangxuan Long, Peiyang Liu, Weidi Xu, Zhe Wang, Shirui Pan, Chao Qu

ACL 2026Findings Findings of the Association for Computational Linguistics: ACL 2026

arXiv

Less Is More: Elevating RAG via Performance-Driven Context Compression

Ziqiang Cui, Yunpeng Weng, Xing Tang, Peiyang Liu, Shiwei Li, Bowei He, Jiamin Chen, Yansen Zhang, Xiuqiang He, Rui Zhang, Chen Ma

ICML 2026CCF-A The 43rd International Conference on Machine Learning

2025

Queries Are Not Alone: Clustering Text Embeddings for Video Search

Peiyang Liu, Xi Wang, Ziqiang Cui, Wei Ye

SIGIR 2025CCF-AOral Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval, pp. 874-883

arXiv

Who Stole Your Data? A Method for Detecting Unauthorized RAG Theft

Peiyang Liu, Ziqiang Cui, Di Liang, Wei Ye

Preprint arXiv:2510.07728, 2025

arXiv

Structural Reward Model: Enhancing Interpretability, Efficiency, and Scalability in Reward Modeling

Xiaoyu Liu, Di Liang, Hongyu Shan, Peiyang Liu, Yonghao Liu, Muling Wu, Yuntao Li, Xianjie Wu, Li Miao, Jiangrong Shen, et al.

EMNLP 2025CCF-B Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: Industry Track, pp. 672-685

Paper

Semantic Retrieval Augmented Contrastive Learning for Sequential Recommendation

Ziqiang Cui, Yunpeng Weng, Xing Tang, Xiaokun Zhang, Shiwei Li, Peiyang Liu, Bowei He, Dugang Liu, Weihong Luo, Xiuqiang He, et al.

NeurIPS 2025CCF-A The 39th Annual Conference on Neural Information Processing Systems

arXiv

CORE-RAG: Lossless Compression for Retrieval-Augmented LLMs via Reinforcement Learning

Ziqiang Cui, Yunpeng Weng, Xing Tang, Peiyang Liu, Shiwei Li, Bowei He, Jiamin Chen, Yansen Zhang, Xiuqiang He, Chen Ma

Preprint arXiv:2508.19282, 2025

arXiv

2024

Unsupervised Corrupt Data Detection for Text Training

Peiyang Liu

ESWA 2024CCF-C Expert Systems with Applications, Volume 248, 123335

Paper

2023

Retrieval-Based Unsupervised Noisy Label Detection on Text Data

Peiyang Liu, Jinyu Yang, Lin Wang, Sen Wang, Yunlai Hao, Huihui Bai

CIKM 2023CCF-B Proceedings of the 32nd ACM International Conference on Information and Knowledge Management, pp. 4099-4104

Paper

2022

Label Smoothing for Text Mining

Peiyang Liu, Xiangyu Xi, Wei Ye, Shikun Zhang

COLING 2022CCF-B Proceedings of the 29th International Conference on Computational Linguistics, pp. 2210-2219

Paper

2021

Improving Embedding-based Large-scale Retrieval via Label Enhancement

Peiyang Liu, Xi Wang, Sen Wang, Wei Ye, Xiangyu Xi, Shikun Zhang

EMNLP 2021Findings Findings of the Association for Computational Linguistics: EMNLP 2021, pp. 133-142

Paper

Distilling Knowledge from BERT into Simple Fully Connected Neural Networks for Efficient Vertical Retrieval

Peiyang Liu, Xi Wang, Lin Wang, Wei Ye, Xiangyu Xi, Shikun Zhang

CIKM 2021CCF-B Proceedings of the 30th ACM International Conference on Information & Knowledge Management, pp. 3965-3975

Paper

QuadrupletBERT: An Efficient Model For Embedding-Based Large-Scale Retrieval

Peiyang Liu, Sen Wang, Xi Wang, Wei Ye, Shikun Zhang

NAACL 2021CCF-B Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 3734-3739

Paper

2020

Not All Synonyms Are Created Equal: Incorporating Similarity of Synonyms to Enhance Word Embeddings

Peiyang Liu, Wei Ye, Xiangyu Xi, Tong Wang, Jinglei Zhang, Shikun Zhang

IJCNN 2020CCF-C 2020 International Joint Conference on Neural Networks, pp. 1-8, IEEE

Paper

Contact联系方式

📧

liupeiyang@pku.edu.cn