I work where AI security, computer vision and vision-language models meet, and I like turning research questions into reproducible tools: every claim should come with the evidence to check it.
- 🛡️ AI & multimodal security · robustness of multimodal AI · adversarial and failure-mode analysis · content integrity · auditable evaluation
- 👁️ Vision & multimodal intelligence · computer vision and VLMs · multimodal reasoning and memory · evaluation under real-world uncertainty · reproducible failure analysis
- ✍️ Preparing submissions to CVPR, ICCV and COLM 2027.
- ⏪ Zerostel, a flight recorder and time machine for AI coding agents, public since 5 October 2026.
- 🚀 Taiwan Exam, an Agent Skill for original GSAT practice exams, public since 28 September 2026.
- 🌙 Built LumiGrid for NTIRE 2025 low-light enhancement: +8.1 dB over the course pipeline it started from.
