Undergraduate in the Arete Honors Program at National Yang Ming Chiao Tung University, majoring in Artificial Intelligence in Engineering and Science.
I work on two sides of one question: as AI systems become more capable and autonomous, how do we make them behave reliably under distribution shift and real-world uncertainty?
🤖 Embodied AI — vision-language-action and world action models at BASIC Lab, advised by Prof. Hong-Han Shuai. Investigating whether a policy's own world-model prediction consistency can serve as a label-free preference signal.
🛡️ AI Safety — at the Institute of Information Science, Academia Sinica, advised by Prof. Lun-Wei Ku. I own the safety track of a trustworthy news LLM built with Taiwan's public media consortium: measuring hallucination, and designing the guardrails and abstention behavior that decide when a model should refuse to answer rather than guess.
MetaCog-Bench · ACL 2026 SRW (Oral) — a diagnostic benchmark showing edge LLMs fabricate plausible tool-call parameters instead of seeking clarification, and that a structured audit prompt collapses hallucination from 32–81% to under 5%.
-
🏆 Meta Wearable AI Challenge @ ECCV 2026 — 2nd and 3rd across three sub-tracks. Every podium entry ran under 5B parameters; the 2B+ result came from a 4.54B model placing above a 27B one.
-
🥈 MindCube Challenge @ CVPR 2026 — 2nd place. Multi-view VLMs for spatial reasoning under partial observability, ~96% on the benchmark.
-
🎖️ CURE-Bench @ NeurIPS 2025 — Excellence Award in Human-Aligned Evaluation. Agent systems for therapeutic reasoning and drug decision-making.
- 🌐 xiaoce.me — publications, research and work experience, and CV
- 📫 hsiao.chun.en@gmail.com
I'm open to internships and research collaborations, and especially curious about under-explored applications of AI — such as AI for archaeology.