Download PDF CVPDF updated 2026-09-28.
Education
National Taiwan University, Graduate Institute of Communication Engineering
Ph.D. at Network Database Laboratory. Supervisor: Prof. Ming-Syan Chen
2021 - Present
National Taipei University of Technology, Graduate Institute of Manufacturing Technology
M.S. at Advanced Microsystems and Device Laboratory. Supervisor: Prof. Chih-Cheng Lu
2018 - 2021
National Taipei University of Technology, College of Mechanical and Electrical Engineering
B.S. in Mechanical Engineering
2014 - 2018
Awards
- 2025 National Taiwan University Wen-Tzu Hsiang Memorial Scholarship
Research Interests
- Efficient Transformer inference: early exiting, KV admission, long-context inference, and adaptive computation for generative retrieval and language models.
- Model compression and quantization: quantization for looped language models and early-exit Vision Transformers, LoRA-guided pruning, and multi-task compression.
- Trustworthy deployment: multi-task unlearning, secure on-device language models with trusted execution environments, and interpretable medical AI.
Skills
Deep Learning
Vision Transformers, Large Language Models, Graph Neural Networks, Convolutional Neural Networks
Model Optimization
Network pruning, quantization, early exiting, LoRA, parameter-efficient fine-tuning
Systems and Deployment
On-device inference, trusted execution environments, CUDA and GPU programming
Programming
Python, C++, CUDA, PyTorch, TensorFlow, JAX
Publications
Mask as Memory: Streaming Multi-task Pruning for Vision Transformers
HSMLog: Small Language Model-Assisted Hardware Security Module Log Anomaly Detection with Behavioral Analysis
LOOPQ: QUANTIZATION FOR LOOPED LANGUAGE MODELS
Interference-Aware Multi-Task Unlearning
Amortized-Precision Quantization for Early-Exit Vision Transformers
LoGIC: Multi-LoRA Guided Importance Consensus for Multi-Task Pruning in Vision Transformers
Dual Alignment Framework for Few-shot Learning with Inter-Set and Intra-Set Shifts
KV Admission: Learning What to Write for Efficient Long-Context Inference
SEAL: Secure and Efficient Adaptive Layering for On-Device Language Models with TEE
BiLEE: Bi-Level Early Exiting for Generative Document Retrieval
Dual-Triangular QR Decomposition with Global Acceleration and Partially Q-Rotation Skipping
A visually interpretable detection method combines 3-D ECG with a multi-VGG neural network for myocardial infarction identification
Professional Activity
- Conference Reviewer: CVPR, ICML, IJCAI, ECAI, ECCV, ICLR, NeurIPS, etc.