About Me

I am Kenneth Yang, an MS student at National Taiwan University advised by Prof. Yu-Chiang Frank Wang. I am currently a Research Intern at the Harvard Kempner Institute, working with Prof. Qianqian Wang on latent action learning.

I received my BS degree in Computer Science from National Tsing Hua University, where I received the Outstanding Academic Achievement Award twice. During my undergraduate studies, I worked with Dr. Jen-Chun Lin on cross-modal modeling and efficient learning.

My research interests lie at the intersection of computer vision and machine learning. My long-term goal is to build intelligent systems that can develop structured representations of the physical world from visual and multimodal experience. I am particularly interested in how models can capture geometry, dynamics, and interactions from observations, and use these representations to reconstruct scenes, predict how the world may evolve, and reason about actions.

📣📣 I plan to apply to CS Ph.D. programs in the Fall 2027 application cycle. I am also actively seeking research internship or visiting student opportunities. Please feel free to reach out by email!

News

Publications

NeurIPS 2026

HIST3R: Rectified State Decomposition from History for Streaming 3D Reconstruction

Kenneth Yang, Yu-Chiang Frank Wang

Conference on Neural Information Processing Systems (NeurIPS), 2026. Poster.

TL;DRPaper and Website coming soon.

TL;DR: A direction–magnitude view of recurrent state updates that unifies and improves CUT3R-based streaming 3D reconstruction.

arXiv 2026

GD-FPS: Growth-Driven Feedforward Parameter Selection for Efficient Fine-Tuning

Kenneth Yang, Wen-Li Wei, Jen-Chun Lin

arXiv preprint, 2026

TL;DRPaper

TL;DR: Select task-sensitive parameters by measuring downstream activation growth against a pre-training anchor, making parameter selection fast, stable, and gradient-free.

NeurIPS COML Workshop 2025

FPS: Feed-Forward Based Parameter Selection for Efficient Fine-Tuning

Kenneth Yang, Wen-Li Wei, Jen-Chun Lin

Conference on Neural Information Processing Systems (NeurIPS), COML Workshop, 2025. Poster.

TL;DRPaper

TL;DR: Select trainable parameters using only forward-pass weight–activation signals, avoiding the full-model backward pass required by gradient-based selection.

ICASSP 2024Oral

Music-to-Dance Poses: Learning to Retrieve Dance Poses from Music

Bo-Wei Tseng, Kenneth Yang, Yu-Hua Hu, Wen-Li Wei, Jen-Chun Lin

IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024. Oral.

TL;DRPaper

TL;DR: A retrieval-based music-to-dance framework that projects music snippets into 3D human pose space.

Education

• National Taiwan University, MS in Communication Engineering (Data Science Group)
  • Advisor: Prof. Yu-Chiang Frank Wang
  • Research: 3D reconstruction, video generation
  • Overall GPA: 4.14/4.3
• National Tsing Hua University, BS in Computer Science
  • Recipient of Outstanding Academic Achievement Award (Top 5% in class), twice
  • Last 60 GPA: 4.24/4.3

Experience

• Research Assistant, Sep 2024 - Present
  Vision and Learning Lab, National Taiwan University
  Advisor: Prof. Yu-Chiang Frank Wang
  • Research areas: 3D reconstruction, video generation
• Summer Research Intern / Research Assistant, Jul 2023 - Sep 2024
  Multimedia Technologies Lab, Academia Sinica IIS
  Advisor: Dr. Jen-Chun Lin
  • Research areas: cross-modal modeling, efficient learning

Teaching

• Graduate Teaching Assistant, Fall 2025
  Deep Learning for Computer Vision, National Taiwan University
  • Graduate-level course for the EECS Department, with more than 250 students enrolled.
  • Held discussion sessions and office hours, designed assignments, graded coursework, and guided course projects.

Honors and Awards

• Outstanding Academic Achievement Award, National Tsing Hua University, Fall 2023
• Outstanding Academic Achievement Award, National Tsing Hua University, Spring 2024