BD Bangde Du

LLM Post-Training · Personalization

杜邦得

Bangde Du

Ph.D. Student, Tsinghua University · Visiting Ph.D. Student, NUS

Singapore

I study how large language models can understand, model, and empower individuals.

Email Google Scholar CV (PDF)
Portrait of Bangde Du

About

About

I am a Ph.D. student in Computer Science at Tsinghua University and a member of THUIR. I am honored to be advised by Prof. Yiqun Liu, and am fortunate to work closely with Prof. Qingyao Ai. I am also a visiting Ph.D. student at the National University of Singapore, hosted by Prof. Chua Tat-Seng.

My work focuses on LLM post-training, personalization, and persona simulation. I am also a research intern at Tencent Hunyuan through Project Up (Qingyun Talent Program), working on post-training for personalized memory.

我是清华大学计算机科学与技术系 THUIR 的博士生。我很荣幸由 刘奕群教授指导攻读博士，并有幸得到 艾清遥教授的指导与帮助。目前，我也在新加坡国立大学访学，合作导师为 Chua Tat-Seng 教授。

我的工作聚焦大语言模型后训练、个性化与人格模拟。我也是腾讯混元青云计划 Project Up (Qingyun Talent Program) 的研究实习生，从事面向个性化记忆的大语言模型后训练。

Experience

Industry Experience

Tencent Hunyuan

May 2026 - Present

Research Intern | Project Up (Qingyun Talent Program)

LLM Post-training for Personalized Memory

  • Personalized memory use. Developed SFT data from teacher-model outputs for task continuity and user-specific interactions, and refined model-based evaluation rubrics. The data was used in Hunyuan post-training.
  • Offline memory selection. Owned training and evaluation of a model for selecting memories to retain during offline updates, covering reward design, full-parameter SFT, and GRPO.
  • 个性化记忆使用。基于教师模型输出构建面向连续任务与用户个体相关交互的 SFT 数据，并完善基于模型的评测准则。相关数据已用于混元后训练。
  • 离线记忆筛选。负责离线更新中筛选保留记忆的模型训练与评测，涵盖奖励设计、全参数 SFT 和 GRPO。

Publications

Papers

See the full and most recent list on Google Scholar.

完整和最新论文列表见 Google Scholar。

First & Co-first Author

SimVBG: Simulating Individual Values by Backstory Generation

Bangde Du, Ziyi Ye, Zhijing Wu, Monika A. Jankowska, Shuqi Zhu, Qingyao Ai, Yujia Zhou, Yiqun Liu

EMNLP 2025, Main Conference First author

Proposed a backstory-based framework for simulating individual values; led its development and evaluation against retrieval-based methods on a benchmark derived from the World Values Survey.

Paper Code

TwinVoice: A Multi-dimensional Benchmark Towards Digital Twins via LLM Persona Simulation

Bangde Du*, Minghao Guo*, Songming He, Ziyi Ye, Xi Zhu, Weihang Su, Shuqi Zhu, Yujia Zhou, Yongfeng Zhang, Qingyao Ai, Yiqun Liu

* Equal contribution.

Findings of ACL 2026 Co-first author

Designed and led the development of a benchmark spanning social, interpersonal, and narrative personas to evaluate LLM persona simulation.

Structured evaluation around six capabilities: memory recall, logical reasoning, opinion consistency, lexical fidelity, persona tone, and syntactic style.

Paper Project Code

Understanding the Effect of Opinion Polarization in Short Video Browsing

Bangde Du, Ziyi Ye, Zhijing Wu, Monika Jankowska, Qingyao Ai, Yiqun Liu

SIGIR 2025

Paper

EEG reveals the cognitive impact of polarized content in short video scenarios

Bangde Du, Ziyi Ye, Monika Jankowska, Zhijing Wu, Qingyao Ai, Yujia Zhou, Yiqun Liu

Scientific Reports, 2025

Paper

Co-authored Publications

Retrievable Gradients: Continual Post-Training Without Cumulative Weight Drift

Weihang Su, Jiacheng Kang, Jingyan Xu, Qingyao Ai, Jianming Long, Hanwen Zhang, Bangde Du, Xinyuan Cao, Min Zhang, Yiqun Liu

EMNLP 2026, Main Conference

Paper

Parametric Social Identity Injection and Diversification in Public Opinion Simulation

Hexi Wang, Yujia Zhou, Bangde Du, Qingyao Ai, Yiqun Liu

KDD 2026

Paper

How do Humans Process AI-generated Hallucination Contents: a Neuroimaging Study

Shuqi Zhu, Yi Zhong, Ziyi Ye, Bangde Du, Yujia Zhou, Qingyao Ai, Yiqun Liu

ICML 2026

Paper

Under Review

Manuscripts Under Review

Mitigating Identity Essentialism in LLM Agents with Longitudinal Life Trajectories

Hexi Wang, Yujia Zhou, Bangde Du, Weihang Su, Xinyuan Cao, Qingyi Pan, Qingyao Ai, Yueyue Wu, Min Zhang, Yiqun Liu

Under review at AAAI 2027

Preprint

Composable Parametric Demographic Injection for Values Survey Simulation

Xinyuan Cao, Qingyi Pan, Bangde Du, Yujia Zhou, Hexi Wang, Weihang Su, Yiqun Liu, Yueyue Wu, Min Zhang, Qingyao Ai

Under review at AAAI 2027

Preprint

News

Recent Updates

Retrievable Gradients was accepted to EMNLP 2026, Main Conference.

Started as a visiting Ph.D. student at the National University of Singapore, hosted by Prof. Chua Tat-Seng.

Joined Tencent Hunyuan as a research intern through Project Up (Qingyun Talent Program), working on post-training for personalized memory.

The hallucination neuroimaging study was accepted to ICML 2026.

The public opinion simulation paper was accepted to KDD 2026.

TwinVoice was accepted to Findings of ACL 2026.

SimVBG was accepted to EMNLP 2025 Main Conference.

The EEG study on polarized short-video content was published in Scientific Reports.

Education

Academic Background

Tsinghua University

Ph.D. in Computer Science

Advisor: Prof. Yiqun Liu

Aug 2023 - 2028 (expected)

National University of Singapore

Visiting Ph.D. Student

Host: Prof. Chua Tat-Seng

Sep 2026 - Present

Tsinghua University

B.Eng. in Automation

Sep 2019 - Jul 2023

Awards

Honors & Scholarships

Tsinghua Friends - Bosch Talent Scholarship, First Class

2025-2026 academic year

University-level Scholarship, Tsinghua University

2022

Second Prize, 40th Tsinghua University Challenge Cup

First author

Contact

Get in touch

dbd23@mails.tsinghua.edu.cn Google Scholar CV (PDF)