Paul (Kuo-Ming) Huang

Currently, I am a research assistant at the Computational Learning Lab at National Taiwan University, advised by Prof. Hsuan-Tien Lin. My research interests mainly lie at the intersection of computer vision and generative AI, especially in improving our controllability over generative AI and the effectiveness of generated data.

I received my undergraduate degree in Computer Science and Information Engineering at National Taiwan University, where I completed my Bachelor's Thesis under the supervision of Prof. Hsuan-Tien Lin. During my undergraduate study, I also had the honor of working with Prof. Hung-Yi Lee.

I was also very fortunate to attend Caltech Summer Undergraduate Research Fellowships (SURF) in 2024, where I was mentored by Prof. Yaser S. Abu-Mostafa at Learning Systems Group.

Email  /  CV  /  Scholar  /  Github

Publications

A survey paper on pitfalls of knowledge editing in large language models. We organized the literatures and benchmarks on the drawbacks of knowledge editing to provide a unified view on this matter.

We demonstrate how parameter-efficient tuning can be applied to large pre-trained sequence-to-sequence speech models and its effectiveness, especially in low-resourced settings.

We demonstrate the potential of classifier guidance in semi-supervised settings and how it can be further improved through the proposed self-calibration technique, which aligns the probability estimation of classifiers with that of score-based models.

We demonstrate the benefit of joint training and classifier calibration to classifier guidance.