2026.08 – Present
Research Intern
UTL Lab · Korea University
Universal Transfer Learning Laboratory, advised by Prof. Donghyun Kim.
Researcher · Engineer · Musician
Seungwoo Kang
I am an undergraduate student in Computer Engineering at Hongik University and an undergraduate researcher at UTL Lab, Korea University. My current research explores offline-to-online reinforcement learning to improve the generalization of vision-language-action (VLA) models. Previously, at SCAI Lab, Chung-Ang University, I worked on knowledge distillation for speech self-supervised learning and training-free steering for large language models. Outside the lab, I compose and produce music.
My research interests include:
- Robot Learning
- Robotic Manipulation
- Large Language Models
Projects
Strategy-Aware Activation Steering for Diverse LLM Decoding
- Designing SAAS, a test-time activation steering framework for strategy-aware diverse decoding
- Developing strategy detection and steering methods based on task-specific activation clusters
- Planning evaluations on math, coding, and creative generation tasks
Jailbreak-Robust ZeroTuning for LLMs
- Advanced Machine Learning Course Project
- Reproduced and evaluated ZeroTuning-based attention sink modulation
- Investigated the relationship between attention modulation and jailbreak robustness
Face-Age Estimation
- Basic Machine Learning Course Project
- Built a face age estimation model using ResNet/VGG backbones on the UTKFace dataset
- Enhanced model generalization performance through data augmentation strategies
Experience & Education
Music
Music & Compositions
Outside of the lab, I spend much of my time playing guitar and experimenting with sound. This is a collection of my original compositions and productions.
* All tracks are self-produced.
What If
feat. HarryChild
0:00
0:00
Tracklist
Contact
I am always happy to talk about research, music, or collaboration.
- Email swkang6348 (at) gmail.com
- GitHub github.com/KSW11390