Hello! I’m a PhD student in the NLP group at King’s College London, working with Prof. Yulan He and Dr. Lin Gui.

My current research focuses on Auto Research, spanning reliable code reproduction from academic papers, the development of LLM-based agents that propose novel scientific ideas and predict promising research directions, and the use of reinforcement learning to enhance these capabilities. I previously worked as a Research Engineer at Meta Superintelligence Labs, where I contributed to the Muse Glimmer and Muse Spark models.

Download CV

News

• Sep. 2026

  Latent Refinement Decoding was accepted to NeurIPS 2026.

• Sep. 2026

  Follow the Latent Roadmap was accepted to EMNLP 2026 as an oral presentation.

• Apr. 2026

  COVER was accepted to ICML 2026.

• Jul. 2025

  SciReplicate-Bench was accepted to COLM 2025.

All news

Experience

Meta — Research Engineer (Nov. 2025 – Sep. 2026)

• Core contributor to Muse Glimmer, an open agentic foundation model.
• Contributed to the development of Muse Spark.
• Developing AI Scientist agents: LLM-based systems that automate scientific research tasks and machine learning engineering workflows.
• Building automated bug-injection pipelines for SWE-bench to construct training and evaluation datasets.
• Building an automated data construction pipeline for recursive self-improvement (RSI) that identifies model-specific weaknesses and generates targeted training data to improve model capabilities.

AstraZeneca — Internship (Jul. 2025 – Oct. 2025)

Supervised by Dr. Saseendran and Dr. Jin. Conducted research on efficient inference for diffusion language models. Published COVER at ICML 2026, achieving up to 11.64× decoding speedup while preserving generation quality.

Publications

• Stop the Flip-Flop: Context-Preserving Verification for Fast Revocable Diffusion Decoding [paper] [Code]
  Yanzheng Xiang, Lan Wei, Yizhen Yao, Qinglin Zhu, Hanqi Yan, Chen Jin, Philip Alexander Teare, Dandan Zhang, Lin Gui, Amrutha Saseendran, Yulan He
  In ICML 2026.
• SciReplicate-Bench: Benchmarking LLMs in Agent-driven Algorithmic Reproduction from Research Papers [arXiv] [Code]
  Yanzheng Xiang, Hanqi Yan, Shuyin Ouyang, Lin Gui, Yulan He
  In COLM 2025.
• G3R: A Graph-Guided Generate-and-Rerank Framework for Cross-domain Text-to-SQL Generation [paper] [Code]
  Yanzheng Xiang, Qian-Wen Zhang, Xu Zhang, Zejie Liu, Yunbo Cao, Deyu Zhou
  In Findings of ACL 2023.
• Addressing Order Sensitivity of In-Context Demonstration Examples in Causal Language Models [arXiv] [Code]
  Yanzheng Xiang, Hanqi Yan, Lin Gui, Yulan He
  In Findings of ACL 2024.
• Latent Refinement Decoding: Enhancing Diffusion-Based Language Models by Refining Belief States [arXiv]
  Qinglin Zhu, Yizhen Yao, Runcong Zhao, Yanzheng Xiang, Amrutha Saseendran, Chen Jin, Philip Teare, Bin Liang, Yulan He, Lin Gui
  In NeurIPS 2026.
• Follow the Latent Roadmap: Navigating Revocable Decoding for Diffusion LLMs with Anchor Tokens [arXiv] [Code]
  Yizhen Yao, Qinglin Zhu, Runcong Zhao, Xiangxiang Dai, Yanzheng Xiang, Yulan He, Lin Gui
  In EMNLP 2026 (Oral).
• Encourage or Inhibit Monosemanticity? Revisit Monosemanticity from a Feature Decorrelation Perspective [paper]
  Hanqi Yan, Yanzheng Xiang, Guangyi Chen, Yifei Wang, Lin Gui, Yulan He
  In EMNLP 2024.
• The Mystery of In-Context Learning: A Comprehensive Survey on Interpretation and Analysis [paper]
  Yuxiang Zhou, Jiazheng Li, Yanzheng Xiang, Hanqi Yan, Lin Gui, Yulan He
  In EMNLP 2024.
• PECAN: LLM-Guided Dynamic Progress Control with Attention-Guided Hierarchical Weighted Graph for Long-Document QA [paper]
  Xinyu Wang, Yanzheng Xiang, Lin Gui, Yulan He
  In Findings of ACL 2025.
• A Divide-And-Conquer Approach for Multi-label Multi-hop Relation Detection in Knowledge Base QA [paper]
  Deyu Zhou, Yanzheng Xiang, Linhai Zhang, Chenchen Ye, Qian-Wen Zhang, Yunbo Cao
  In Findings of EMNLP 2021.