JCW_
ONLINE

~/john-wei · ntu.eecs

Jui-Chiang
Wei

魏睿強 · M.S. @ NTU · Efficient AI systems, model compression & hardware-aware ML

I am a master's student in Electronics Engineering at National Taiwan University, advised by Prof. An-Yeu (Andy) Wu in the Access IC Design Lab. I build efficient AI systems across post-training quantization, Mamba-based models, and accelerator-aware optimization.

// 02 trajectory 04 nodes

NTU emblem
CURRENT

NTU

M.S. Student · Graduate Institute of Electronics Engineering, ICS Group

Academia Sinica emblem
INTERN

Academia Sinica

Research Intern · TinyML, sustainable AI & efficient deployment

NTU emblem
UNDERGRAD

NTU

B.S. Student · Department of Electrical Engineering

NCTU emblem
UNDERGRAD

NCTU

B.S. Student · Department of Electrical Engineering

// 03 publications 04 entries

Research focus

My work asks how we can make advanced models more practical: smaller, faster, and aware of the systems that run them. I am especially interested in efficient vision models, post-training quantization, Mamba architectures, and AI accelerators.

model compressionquantizationMambaIC designvision modelsefficient AI

ICIP · 2026 NEW

HyQuant: A Unified Quantization Framework for Hybrid Mamba–Transformer Vision Models

Jui-Chiang Wei, B.Y. Shi, An-Yeu Wu

A systematic post-training quantization framework for hybrid vision architectures including MambaVision and VSSD.

ASRU · 2025

Multi-Distillation from Speech and Music Representation Models

Jui-Chiang Wei, Yi-Cheng Lin, Fabian Ritter-Gutierrez, Hung-yi Lee

A multi-teacher framework for unifying speech and music representations in one compact model.

Interspeech · 2025

Distilling a Speech and Music Encoder with Task Arithmetic

Fabian Ritter-Gutierrez, Yi-Cheng Lin, Jui-Chiang Wei, et al.

A task-arithmetic distillation approach for combining independent speech and music encoders.

ICLR · 2024

Dynamic-SUPERB Phase-2

Chien-yu Huang, Wei-Chih Chen, Shu-wen Yang, et al.

A collaboratively expanding benchmark for measuring spoken-language models across 180 tasks.

// 04 selected projects 03 entries

Multi-Core GEMM Accelerator

A multi-core accelerator with distributed memory, AXI data movement, and controller-managed scheduling across processing elements.

AcceleratorRTLAXI

ViT Patch Embedding Accelerator

A Vision Transformer patch-embedding accelerator with INT8 quantization, memory access, and clock-domain-crossing synchronization.

VisionINT8Hardware

BCH Decoder

A BCH decoder supporting hard- and soft-decision decoding, Berlekamp error-location computation, and Chase decoding.

Coding TheoryDigital Design

// 05 contact

Let's work on
efficient intelligence.

johnwei@access.ee.ntu.edu.tw