Virginia Tech · Computer Science
Xinyue Zeng (Susan)
I am a Ph.D. candidate in Computer Science at Virginia Tech, advised by Prof. Dawei Zhou. My research focuses on self-improving AI agents and reliable, efficient LLM reasoning, with an emphasis on learning from experience, RL post-training, and test-time scaling. I am also exploring recursive self-improvement (RSI), particularly how agents can improve the way they learn and adapt.
In Summer 2026, I was invited to join Microsoft’s Frontier Tuning AI team as a research intern, extending my research on test-time inference and hallucination control to long-horizon AI agents. My statistical training spans a B.S. in Statistics from Zhejiang University and an M.S. in Biostatistics from UNC–Chapel Hill.
Recent updates
-
My first-author paper, ForkPilot, developed during my internship with Microsoft’s Frontier Tuning AI team, is now available on arXiv.
-
SAGE was accepted to the NeurIPS 2026 main conference.
-
I was invited to join Microsoft’s Frontier Tuning AI team as a research intern, extending my research on test-time inference and hallucination control.
-
Received the ICLR 2026 Tier One Financial Travel Award for my two accepted papers.
-
Featured in the Sanghani Center Student Spotlight and received an ACM CAPWIC Travel Scholarship.
-
HalluGuard and Plan-and-Budget were accepted to ICLR 2026; research on foundational LLM reasoning was accepted to ACM CAPWIC 2026.
-
EPIC-unmix was published in Genome Biology.
Older updates
-
Research on foundational LLM fine-tuning was accepted to the ICDM 2025 Ph.D. Forum.
-
LENSLLM was accepted to ICML 2025.
Contact
Research & collaboration
I welcome collaborations in self-improving agents, reliable and efficient reasoning, and AI for science.
Virginia Tech · Blacksburg, VA