Rui Fang

Rui Fang

Ph.D. Student@GICE, NTU

Personal Page

I am a Ph.D. candidate at the National Taiwan University (NTU) in the Graduate Institute of Communication Engineering, under the supervision of Prof. Ming-Syan Chen. My research is centered on efficient deep learning for Transformers, with broader interests spanning few-shot learning and interpretable machine learning. I am committed to developing scalable and efficient AI models that address real-world problems while maintaining model interpretability.

Research Highlights

Efficient Inference for Transformers (2024–Present)

My recent work focuses on enabling scalable, low-latency inference without sacrificing performance:

  • “BiLEE: Bi-Level Early Exiting for Generative Document Retrieval” (ECAI 2024) introduces a bi-level early exiting strategy that significantly accelerates document retrieval without accuracy loss
  • “LoGIC: Multi-LoRA Guided Importance Consensus for Multi-Task Pruning in Vision Transformers” (AAAI 2026) proposes a novel pruning framework that leverages LoRA for efficient multi-task learning
  • “Learning What to Write: Write-Gated KV for Efficient Long-Context Inference” (arXiv 2025) addresses KV-cache compression for long-context inference in LLMs

Few-Shot Learning under Distribution Shifts (2023–2024)

I collaborated on “Dual Alignment Framework for Few-shot Learning with Inter-Set and Intra-Set Shifts” (NeurIPS 2025), which tackles the dual domain adaptation challenge in few-shot learning scenarios through optimal transport-based feature calibration.

Interpretable ECG Diagnosis (2019–2021)

My early work on “A Visually Interpretable Detection Method Combines 3-D ECG with a Multi-VGG Neural Network for Myocardial Infarction Identification” (Computer Methods and Programs in Biomedicine 2022) demonstrated the importance of interpretability in medical AI applications, combining 3D electrocardiogram data with deep learning for explainable myocardial infarction detection.

I am dedicated to advancing the field of AI through research that is both technically sound and practically applicable, with the ultimate goal of creating innovations that serve societal needs.