AI Security ×ばつ Trustworthy Evaluation ×ばつ Evidence-Grounded Systems
Researching the security, robustness, and reasoning capabilities of visual and multimodal intelligent systems.
Currently taking a one-year leave of absence from the master's program.
I earned my degree in Computer Science from Yuan Ze University (YZU) and am now enrolled in a master's program at National Yang Ming Chiao Tung University (NYCU). I am currently taking a one-year leave of absence. My primary interests lie at the intersection of AI security, computer vision, and vision-language models, with an emphasis on reliable multimodal reasoning and evaluation.
I turn research questions into reproducible experiments and working systems. Current projects include KCrashLab, a deterministic platform for reproducible Windows driver reliability research, and ContextSec, a product-security decision layer for AI coding agents.
- Security and robustness of multimodal AI
- Adversarial and failure-mode analysis
- Content integrity and verification
- Trustworthy, auditable evaluation
- Computer vision and vision-language models
- Multimodal reasoning and memory
- Evaluation under real-world uncertainty
- Reproducible failure analysis
ContextSec
A research-preview, deterministic product-security decision layer for AI coding agents. It derives applicable risk packs from bounded repository evidence, composes cross-context controls, and emits a verifiable Control Evaluation Ledger with an explicit release gate. Merriv release evidence architecture
Merriv
A pre-alpha, vendor-neutral release-evidence layer for deployable AI models. It binds exact artifacts, paired evaluation, statistical policy, provenance, and regression onset into a portable Model Change Report for independent verification before promotion.
KCrashLab
Evidence-frozen research platform for deterministic Windows driver reliability experiments. It combines canonical cases, resumable campaigns, exact failure signatures, minimization, 3/3 simulated replay, and semantic artifact verification while clearly separating simulated evidence from the gated Windows-lab track. ChromaRecover evidence-first computer vision architecture
ChromaRecover
An experimental public-alpha, local-first Python toolkit for recovering spatial structure carried by subtle color differences. It tests competing chromatic hypotheses, preserves auditable artifacts, and abstains when evidence is weak.
Python · PyTorch · OpenCV · scikit-learn · Jupyter · C# / .NET · React · TypeScript · Node.js · SQLite · Git
I welcome thoughtful conversations about AI security, CV / VLM reasoning, trustworthy machine learning, reproducibility, and research tooling. Feel free to contact me at wilbur930202@gmail.com .