Rui Fang

Efficient deep learning, from model compression to trustworthy deployment.

Rui Fang

I am a Ph.D. candidate in the Graduate Institute of Communication Engineering at National Taiwan University, advised by Prof. Ming-Syan Chen. My research focuses on making Transformer-based systems faster, smaller, and more deployable without losing the behavior that makes them useful.

Research Interests

Efficient Transformer Inference

Early exiting, KV admission, long-context inference, and adaptive computation for practical latency and memory reduction.

Compression and Quantization

Quantization, pruning, and task-aware model compression for Vision Transformers and recursive language models.

Trustworthy Deployment

Multi-task unlearning, TEE-aware on-device language models, and methods that preserve reliability under deployment constraints.

Recent Work

LoopQ studies post-training quantization for recursive Transformers and introduces loop-aware adaptations for stable low-bit inference.
Amortized-Precision Quantization formulates precision allocation for early-exit Vision Transformers under dynamic inference paths.
KV Admission learns what to write into the KV cache for efficient long-context inference.
LoGIC uses Multi-LoRA guided importance consensus for multi-task pruning in Vision Transformers.
BiLEE introduces bi-level early exiting for generative document retrieval.

Publications

My complete publication list is maintained on the Publications page and mirrored with my Google Scholar profile.

Contact

I am interested in efficient AI systems, model compression, long-context inference, and deployable machine learning. The fastest way to reach me is by email at rfang@arbor.ee.ntu.edu.tw.