Reasoning
Adaptive inference, formal verification, and agentic systems that allocate effort where it matters.
AI Researcher · Tencent
Yan (Yana) Xu 徐妍
Building reliable, reasoning-capable foundation models.
I am an AI researcher at Tencent, working across foundation model reasoning, knowledge grounding, and reliable AI systems. I received my Ph.D. in ECE from HKUST, advised by Prof. Pascale Fung.
15.6k+ Google Scholar citations
20 h-index
2026 Latest work · ACL & agents
Research focus
From grounded answers to adaptive reasoning.
My work studies how foundation models can reason efficiently, use knowledge and tools faithfully, and become more dependable systems in the real world.
Knowledge & tools
Methods that connect models to evidence, retrieval, and tools while preserving usefulness and factuality.
Reliability
Evaluation and mitigation of hallucination, robustness, and controllability in foundation model systems.
Selected work
Recent research.
A selection from current work in adaptive reasoning, proactive agents, and reliable foundation models.
Y.
Blog
A space to think in public—about foundation models, research, technology, and whatever else feels worth writing down.
Browse all posts First post coming soon. This space will collect essays, opinions, and shorter notes as they are published.
Now & next
Updates.
Recent milestones from my research and academic work.
Discover and Prove appeared at ACL 2026.
Released BALTO, a token-level policy optimization framework for hallucination mitigation.
QFFT was accepted as a NeurIPS 2025 Spotlight.
Completed my Ph.D. thesis, Unstructured Knowledge Grounding for Open-Domain Dialogues.
Scholar metrics shown above were checked on August 27, 2026.