Xuan (Lily) Yang 杨萱

I am a PhD student in Computer Science at Duke University, advised by Jian Pei, with expected graduation in 2027.

I study data-centric AI: how to value and curate data, and how to use it more effectively in learning and reasoning. My work spans data valuation, LLM reasoning, and agent evaluation. I am currently exploring data curation for LLMs and agent memory.

Previously, I received my bachelor’s and master’s degrees from Zhejiang University, where I worked with Yang Yang. I have interned at Google and TikTok, and visited Stanford and NUS as a research intern.

I welcome research collaborations and conversations about academic or industry research opportunities. Reach me at xuan.yang@duke.edu.

CV (PDF) xuan.yang@duke.edu Google Scholar LinkedIn

News

Jul 2026Batch-of-Thought accepted to EMNLP 2026.
Jul 2026Local Shapley accepted to VLDB 2026.
Jun 2026MPCEval accepted to KDD 2026 as an oral presentation.

More updates

Jun 2026We released TUA-Bench, a benchmark for terminal-use agents.
May 2026Joined Google Cloud as a summer intern.
Apr 2026Completed my research internship at TikTok.

Selected publications

Full list on Google Scholar

Method overview from the Local Shapley paper

VLDB 2026Data valuation

Local Shapley: Model-Induced Locality and Optimal Reuse in Data Valuation

Xuan Yang, Hsi-Wen Chen, Ming-Syan Chen, Jian Pei

Uses model-induced locality to focus valuation on influential data subsets. LSMR achieves the minimum number of retraining operations for exact Local Shapley computation.

Paper Code

Method overview from the Batch-of-Thought paper

EMNLP 2026LLM reasoning

Batch-of-Thought: Cross-Instance Learning for Enhanced LLM Reasoning

Xuan Yang, Furong Jia, Roy Xie, Xiong Xi, Hengwei Bian, Jian Li, Monica Agrawal

Reasons across related queries to share useful patterns and catch errors. BoT-R improves accuracy and calibration while reducing inference costs by up to 61% across the evaluated benchmarks.

Paper Code

Preprint · 2026Dynamic data valuation

Dynamic Shapley Computation

Xuan Yang, Hsi-Wen Chen, Ming-Syan Chen, Jian Pei

Maintains data values as training data and tasks change. D-Shap supports millisecond task updates and reduces player-update costs by up to three orders of magnitude in experiments.

Paper

Agent evaluation & benchmarks

KDD 2026Oral presentation

MPCEval: A Benchmark for Multi-Party Conversation Generation

Minxing Zhang, Yi Yang, Zhuofan Jia, Xuan Yang, Jian Pei, Yuchen Zang, Xingwang Deng, Xianglong Chen

Evaluates who speaks, what they say, and whether the two are consistent, for both next-turn prediction and full-conversation generation.

Paper

Preprint · 2026Benchmark

TUA-Bench: A Benchmark for Terminal-Use Agents

Shoufa Chen, Luyuan Wang, Xuan Yang, Zhiheng Liu, Yuren Cong, Yuanfeng Ji, Feiyan Zhou, Xiaohui Zhang, Fanny Yang, Belinda Zeng

120 real-world tasks across five task families, from document editing and email to scientific workflows, with reproducible setup and execution-based evaluation.

Project Paper Code

Benchmark overview

Graph learning & applications

AAAI 2023Distinguished
Paper Award

DropMessage: Unifying Random Dropping for Graph Neural Networks

Taoran Fang, Zhiqing Xiao, Chunping Wang, Jiarong Xu, Xuan Yang, Yang Yang

Unifies random dropping in GNNs at the message level, improving training stability and preserving information diversity.

Paper

Method overview

Overview of message dropping in graph neural networks

Scientific Reports2023

Unfolding and Modeling the Recovery Process after COVID Lockdowns

Xuan Yang, Yang Yang, Chenhao Tan, Yinghe Lin, Zhengzhe Fu, Fei Wu, Yueting Zhuang

Uses electricity data and graph learning to model economic recovery and dependencies across sectors. Featured on the front page of Zhejiang Daily.

Paper Press coverage

Research overview

Visualization of the economic recovery research

IEEE TKDE2022

Who’s Next: Rising Star Prediction via Diffusion of User Interest in Social Networks

Xuan Yang, Yang Yang, Jintao Su, Yifei Sun, Shen Fan, Zhongyao Wang, Jun Zhang, Jingmin Chen

Predicts emerging popular items by modeling how user interests spread through social networks, evaluated on large-scale Taobao data.

Paper arXiv

Method overview

Overview of the RiseNet prediction framework

Experience & education

Research experience

Google
May – Aug 2026
PhD Intern · Google Cloud
Kirkland, WA
Built multi-agent systems for performance optimization and studied efficient embedding model distillation.
TikTok
May 2025 – Apr 2026
Research Intern · Risk Control
Bellevue, WA
Deployed multi-agent risk workflows for E-commerce, increasing pre-GMV fraud recall by 26% while reducing manual review effort by 96%.
Alibaba Group
Oct 2020 – Dec 2021
Research Intern · Data Assets and Algorithm
Hangzhou, China
Built large-scale recommendation data and deployed Taobao ranking models, improving CTR by 2.3% over the production baseline.
Stanford University
Jan – Mar 2019
Research Assistant · Center for Magnetic Nanotechnology
Palo Alto, CA
National University of Singapore
Jun – Aug 2018
Research Assistant · Big Brain, BIGHEART
Singapore

Education

2023 – Present · Expected 2027
Duke University
PhD in Computer Science
Advisor: Jian Pei
2020 – 2023
Zhejiang University
MS in Computer Science
2016 – 2020
Zhejiang University
BS in Digital Media Technology

Xuan (Lily) Yang 杨萱

News

Selected publications

Local Shapley: Model-Induced Locality and Optimal Reuse in Data Valuation

Batch-of-Thought: Cross-Instance Learning for Enhanced LLM Reasoning

Dynamic Shapley Computation

Agent evaluation & benchmarks

MPCEval: A Benchmark for Multi-Party Conversation Generation

TUA-Bench: A Benchmark for Terminal-Use Agents

Graph learning & applications

DropMessage: Unifying Random Dropping for Graph Neural Networks

Unfolding and Modeling the Recovery Process after COVID Lockdowns

Who’s Next: Rising Star Prediction via Diffusion of User Interest in Social Networks

Experience & education

Research experience

Google

TikTok

Alibaba Group

Stanford University

National University of Singapore

Education

Duke University

Zhejiang University

Zhejiang University

Teaching & academic service

Teaching assistant · Duke University

Applied Machine Learning and Data Science

Databases

Reviewer

NeurIPS 2026

ACM Transactions on Knowledge Discovery from Data