Xingchen Wan

Research Scientist

Google

About me

I am a Research Scientist at Google based in the San Francisco Bay Area.

I did my DPhil (the Oxford way of saying PhD) in the Machine Learning Research Group, Department of Engineering Science, University of Oxford. I was also a Clarendon Scholar and a member of St John’s College, both at the University of Oxford. I previously interned at Google and Meta.

Academic Services

Reviewer/program committee member at ACL (2023-24), AutoML-Conf (2023-24), COLM (2024), CVPR (2024), ECCV (2024), EMNLP (2023-24), ICLR (2024-25), ICML (2023-24), JMLR, Machine Learning, NeurIPS (2022-23), WACV (2022-24), etc.

Area chair/senior program committee member at NeurIPS (2024), ICML (2025); Action editor at TMLR.

Interests

Large language models
Bayesian optimization
Automated machine learning (AutoML)
Machine learning on graphs

Education

DPhil (PhD), Machine Learning, 2019 - 2023

University of Oxford
BA, MEng, Engineering Science. First-class Honours, 2015 - 2019

University of Oxford

Recent News

💡 Preprint: Automating multi-agent designs

New paper Feb 2025

Our new work on automating prompt and topology design in multi-agent LLM systems is now available as an arXiv preprint!

📄 ICLR acceptance: Self-improving long-context reasoners

New acceptance Feb 2025

Our work, From Few to Many: Self-Improving Many-Shot Reasoners Through Iterative Optimization and Generation, was accepted at ICLR 2025.

💡 Preprint: Astute RAG

New paper Oct 2024

Presenting AstuteRAG, our new work on overcoming imperfect retrieval and knowledge conflict.

📄 3 papers accepted at NeurIPS 2024 + 1 paper accepted at EMNLP 2024!

New acceptance Sep 2024

A comprehensive study of instructions and demonstrations in automatic prompt design, which also powers the recently launched Vertex AI Prompt Optimizer [Google Cloud Blog] [Google for Developers Blog] [Vertex AI Documentation];
Unstructured Query Engine (UQE), an LLM-powered database engine on unstructured data;
GraphComBO [code], a novel Bayesian optimization framework operating over node subsets in graphs.
ZePO(code) which aligns LLM-as-a-judge with human preferences with intelligent prompting was accepted at EMNLP 2024 as a main paper.

Featured

From Few to Many: Self-Improving Many-Shot Reasoners Through Iterative Optimization and Generation

Self-improving reasoners taking advantage of long-context capabilities (ICLR 2025)

Multi-Agent Design: Optimizing Agents with Better Prompts and Topologies

Automatically discovering performant multi-agent systems end-to-end. (arXiv preprint)

Teach Better or Show Smarter? On Instructions and Exemplars in Automatic Prompt Optimization

Comprehensive study powering the Google Cloud Vertex AI Prompt Optimizer (NeurIPS 2024).

Astute RAG: Overcoming Imperfect Retrieval Augmentation and Knowledge Conflicts for Large Language Models

Overcoming RAG imperfections and knowledge conflicts with debating and reconcilation (arXiv preprint).

Publications

View all listed publications or view by tags. For a complete list including preprints & working papers, refer to my Google Scholar.

Xingchen Wan, Han Zhou, Ruoxi Sun, Hootan Nakhost, Ke Jiang, Sercan Ö. Arık (2025). From Few to Many: Self-Improving Many-Shot Reasoners Through Iterative Optimization and Generation. International Conference on Learning Representations (ICLR).

PDF Cite Abstract

Han Zhou, Xingchen Wan, Ruoxi Sun, Hamid Palangi, Shariq Iqbal, Ivan Vulić, Anna Korhonen, Sercan Ö. Arık (2025). Multi-Agent Design: Optimizing Agents with Better Prompts and Topologies. arXiv preprint arXiv:2502.02533.

PDF Cite Abstract

Xingchen Wan, Ruoxi Sun, Hootan Nakhost, Sercan Ö. Arık (2024). Teach Better or Show Smarter? On Instructions and Exemplars in Automatic Prompt Optimization. Advances in Neural Information Processing Systems (NeurIPS).

PDF Cite Abstract OpenReview

Hanjun Dai, Bethany Yixin Wang, Xingchen Wan, Bo Dai, Sherry Yang, Azade Nova, Pengcheng Yin, Phitchaya Mangpo Phothilimthana, Charles Sutton, Dale Schuurmans (2024). UQE: A Query Engine for Unstructured Databases. Advances in Neural Information Processing Systems (NeurIPS).

PDF Cite Abstract OpenReview

Huidong Liang, Xingchen Wan, Xiaowen Dong (2024). Bayesian Optimization of Functions over Node Subsets in Graphs. Advances in Neural Information Processing Systems (NeurIPS).

PDF Cite Code Abstract OpenReview

Han Zhou, Xingchen Wan, Yinhong Liu, Nigel Collier, Ivan Vulić, Anna Korhonen (2024). Fairer Preferences Elicit Improved Human-Aligned Large Language Model Judgments. Empirical Methods in Natural Language Processing (EMNLP).

PDF Cite Code Abstract

Fei Wang, Xingchen Wan, Ruoxi Sun, Jiefeng Chen, Sercan Ö. Arık (2024). Astute RAG: Overcoming Imperfect Retrieval Augmentation and Knowledge Conflicts for Large Language Models. arXiv preprint arXiv:2410.07176.

PDF Cite Abstract

Han Zhou, Xingchen Wan, Lev Proleev, Diana Mincu, Jilin Chen, Katherine Heller, Subhrajit Roy (2024). Batch Calibration: Rethinking Calibration for In-Context Learning and Prompt Engineering. International Conference on Learning Representations (ICLR).

PDF Cite Google Research Blog Abstract OpenReview

Masaki Adachi, Satoshi Hayakawa, Martin Jørgensen, Xingchen Wan, Vu Nguyen, Harald Oberhauser, Michael Osborne (2024). Adaptive Batch Sizes for Active Learning: A Probabilistic Numerics Approach. International Conference on Artificial Intelligence and Statistics (AISTATS).

PDF Cite Code Abstract

Dongyu Gong, Xingchen Wan, Dingmin Wang (2024). Working Memory Capacity of ChatGPT: An Empirical Study. AAAI Conference on Artificial Intelligence (AAAI).

PDF Cite Code Abstract

See all publications