LLM · Agent · Recommendation & Ads · AI for Society

Yu Lei

I study how large language models learn from context, retrieve evidence, compress information, make decisions, and align with human values in realistic tasks.

My work bridges research systems and industrial algorithms, spanning LLM agents, context learning, long-context compression, recommender systems, ad bidding, financial AI, and AI for society.

Research & Publications Google Scholar

About

I work at the intersection of research questions and deployed systems, turning ideas into experiments, evaluation tools, automation workflows, and product-facing models.

I am especially interested in reliable decision-making under incomplete information, long-horizon tasks, and real operational constraints. Across RAG, context compression, recommendation, advertising, financial risk modeling, and value alignment, the common question is how to help models reason with limited evidence and competing objectives.

I received my academic master's degree in Intelligent Science and Technology from the School of Artificial Intelligence, Beijing University of Posts and Telecommunications.

10+papers and submissions
Multipledeployed strategies and systems

Research & Publications

LLM Agents, Context Learning, and Long-Context Compression

I study how large language models use external context: planning and evidence management for deep research, learning reusable skills from long contexts, and compressing information while preserving factual memory.

  • Built agentic research systems and automation workflows for evidence retrieval, context control, report writing, experiment debugging, and knowledge capture.
  • Investigated context-to-skill learning and faithful context compression as complementary ways to make long-context reasoning more reliable.
Agentic RAG Context Learning Context Compression
Tsinghua NLP LabKuaishou
Representative papers
  1. Co-firstNeurIPS 2026 Under Review

    From Context to Skills: Can Language Models Learn from Context Skillfully?

    Shuzheng Si*, Haozhe Zhao*, Yu Lei*, Qingyi Wang*, Dingwei Chen, Zhitong Wang, Zhenhailong Wang, Kangyang Luo, Zheng Wang, Gang Chen, Fanchao Qi, Minjia Zhang, Maosong Sun.

  2. Co-firstEMNLP 2026 Under Review

    From Context to EDUs: Faithful and Structured Context Compression via Elementary Discourse Unit Decomposition

    Yiqing Zhou*, Yu Lei*, Shuzheng Si*, Qingyan Sun, Wei Wang, Yifei Wu, Hao Wen, Gang Chen, Fanchao Qi, Maosong Sun.

  3. First AuthorEMNLP 2026 Under Review

    RhinoInsight: Improving Deep Research through Control Mechanisms for Model Behavior and Context

    Yu Lei, Shuzheng Si, Wei Wang, Yifei Wu, Gang Chen, Fanchao Qi, Maosong Sun.

Generative Recommendation, Ad Bidding, and Financial AI

I work on decision systems under resource and objective constraints, including cold-start content allocation, automated ad bidding, and multimodal financial risk modeling.

  • Explored user-to-photo recommendation and cold-start exposure allocation, connecting platform resource constraints with generative recommendation objectives.
  • Built deployed ad bidding and financial AI systems, including generative bidding, offline bidding simulation, multimodal risk prediction, and financial credit LLMs.
Generative Recommendation Ad Bidding Financial LLM
KuaishouMeituan Waimai AdsDiDi
Representative papers
  1. First AuthorKDD 2026

    Generative Large-Scale Pre-trained Models for Automated Ad Bidding Optimization

    Yu Lei, Jiayang Zhao, Yilei Zhao, Zhaoqi Zhang, Linyou Cai, Qianlong Xie, Xingxing Wang.

  2. First AuthorACL 2026 Industry Track Oral

    A Unified Framework for Modeling Heterogeneous Financial Data via Dual-Granularity Prompting

    Yu Lei, Zixuan Wang, Yiqing Feng, Junru Zhang, Yahui Li, Chu Liu, Tongyao Wang.

  3. First AuthorICDE 2025 Workshop

    Zigong 1.0: A large language model for financial credit

    Yu Lei, Zixuan Wang, Chu Liu, Tongyao Wang.

AI for Society, Safety, and Brain-LLM Alignment

I use neuroscience and social science perspectives to understand LLM representations, behavioral evolution, value alignment, and human-like responses in social settings.

  • Compared LLM embeddings with human fMRI responses to study brain-like hierarchical representations in language models.
  • Studied socially adaptive belief evolution, fairness enforcement, prosocial behavior, and moral decision-making through LLM experiments and agent simulations.
LLM Evaluation Value Alignment Agent Simulation
BUPTTsinghua School of Social SciencesLMU MunichOxford
Representative papers
  1. First AuthorKDD 2026 Oral

    Are LLMs Socially Adaptive? Contrasting Belief Evolution in Large Language Models and Humans

    Yu Lei, Hao Liu, Chengxing Xie, Songjia Liu, Zhiyu Yin, Canyu Chen, Guohao Li, Philip Torr, Zhen Wu.

  2. First AuthorAAAI 2026
  3. Co-firstScience China Tech. Sci.

    Prosocial behavior in Large Language Models: Value alignment and affective mechanisms

    Hao Liu*, Yu Lei*, Zhen Wu.

  4. ContributorNature Human Behaviour Under Review

    Outraged AI: Large language models prioritise emotion over cost in fairness enforcement

    Hao Liu, Yiqing Dai, Haotian Tan, Yu Lei, Yujia Zhou, Zhen Wu.

Contact

Feel free to reach out by email or visit Google Scholar for a fuller publication record.

Email leiyu0210@gmail.com Scholar Google Scholar Homepage leiyu0210.github.io