Kun-Hsiang Lin

About Me

With over 7 years of experience as a CV/AI engineer and researcher, I am currently pursuing a Ph.D. at National Taiwan University, part of the AIMM group in the Communication and Multimedia Lab (CMLab), advised by Prof. Wen-Huang Cheng. My academic journey began with traditional machine learning techniques like support vector machines and random forests for rainfall forecasting during my master’s. I later worked at Academia Sinica on spatial-temporal modeling with deep learning and supported paper reviews for top-tier ML conferences.

After completing my research at Academia Sinica, I spent over five years in industry at Authme as an AI Team Lead and CV/ML Engineer. I led the development of end-to-end AI solutions for biometric security, OCR, and fraud detection—covering algorithm design, C++ SDKs, cloud infrastructure, and MLOps. My work earned top rankings in international challenges such as CVPR, ICCV, and NIST FRTE.

Currently, my research focuses on vision-language models (VLMs) and multimodal large language models (MLLMs). I’m particularly interested in leveraging instruction tuning and multimodal reasoning to improve generalization and interpretability in real-world AI applications.

For more information, please download my CV.

Research Interests

• Computer Vision: Image recognition, object detection, image segmentation, video understanding, and anomaly detection.
• Deep Learing: Domain generalization, self-supervised learning, vision-language models (VLMs), multimodal large language models (MLLMs), image-text matching, multimodal pretraining, knowledge distillation, model compression, and transfer learning.
• Biometric AI: Face recognition, face anti-spoofing, fraud detection, and deepfake detection.

News

• [Jul. 2025] Our paper on face anti-spoofing with vision-language models (VLMs) has been accepted to ACM Multimedia 2025.

Publications

Projects

Powered by Jekyll and Minimal Light theme.