Nguyen Quy Dang
AI research & machine learning systems

Nguyen Quy
Dang.

Making intelligent systems
more efficient.

I’m a Computer Science master’s student at MBZUAI, exploring the intersection of machine learning, compression, and high-performance computing.

Explore my research Download CV

Abu Dhabi, UAE M.Sc. at MBZUAI

01

High-performance computing

02

Systems for machine learning

03

Machine learning for systems

A little background

From algorithms to systems.

I’m interested in how algorithmic ideas translate into efficient, practical machine learning systems. My current research at MBZUAI focuses on efficient machine learning systems and high-performance computing.

Previously, I worked on GPU-based compression and streaming algorithms at BioTuring, and optimized model training and inference on NVIDIA and AMD accelerators at MOREH. My research at AVITECH includes robust tensor decomposition, with two papers at ICASSP 2026.

Across research and engineering, my work spans GPU kernels, sparse-data compression, streaming algorithms, and federated learning. I enjoy connecting mathematical structure with the realities of memory, computation, and scale.

Selected research

Publications.

Download bibliography
2026
ICASSP · FIRST AUTHOR

RE-LL1: An Effective Regularized (L, L, 1)-Tensor Decomposition Method for Video Background Modeling and Foreground Separation

Nguyen Quy Dang, Thanh Trung Le, Nguyen Linh Trung, Karim Abed-Meraim

IEEE International Conference on Acoustics, Speech and Signal Processing

Paper ↗PDF ↗Code ↗
BibTeX
@inproceedings{nguyen2026rell1,
  author = {Nguyen, Quy Dang and Le, Thanh Trung and Nguyen, Linh Trung and Abed-Meraim, Karim},
  title = {{RE-LL1}: An Effective Regularized {(L,L,1)}-Tensor Decomposition Method for Video Background Modeling and Foreground Separation},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  year = {2026},
  url = {https://ieeexplore.ieee.org/document/11460481}
}
2026
ICASSP · CO-FIRST AUTHOR

Fast and Robust Triple Tensor Decomposition With Data Corruption

Nguyen Quy Dang*, Do Minh Nhat*, Thanh Trung Le, Nguyen Linh Trung, Karim Abed-Meraim

IEEE International Conference on Acoustics, Speech and Signal Processing · * Equal contribution

Paper ↗PDF ↗Code ↗
BibTeX
@inproceedings{nguyen2026robusttriple,
  author = {Nguyen, Quy Dang and Do, Minh Nhat and Le, Thanh Trung and Nguyen, Linh Trung and Abed-Meraim, Karim},
  title = {Fast and Robust Triple Tensor Decomposition With Data Corruption},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  year = {2026},
  note = {Nguyen Quy Dang and Do Minh Nhat are co-first authors},
  url = {https://ieeexplore.ieee.org/document/11463179}
}

Ideas into implementation

Selected projects.

GPU systems01

StyleGAN performance
optimization

Optimized training on an RTX 3090 through operator fusion, shared-memory tiling, mixed precision, and overlapping computation with data loading.

CUDA / PyTorch / RTX 3090Training systems
GPU systems02

GPT-OSS inference
optimization

Accelerated inference on AMD MI250 GPUs with attention kernels, contiguous KV-cache layouts, fused operations, and multi-GPU parallelism.

HIP / ROCm / AMD MI250Inference systems
Distributed learning03

Federated learning under
data heterogeneity

Compared FedAvg, FedProx, and FedNova on non-IID MRI data. Built a simulation and visualization platform to study model performance across client distributions.

PyTorch / ViT / Statistical testingGitHub ↗
Multimodal AI04

Visual Language
DENSO Expert

A factory-support system combining vision-language reasoning, document retrieval, speech interaction, and synthetic data. Second Runner-up at DENSO Hackathon 2024.

GPT-4o / Whisper / DiffusionHackathon project
Reinforcement learning05

Multi-agent self-play

Designed convolutional policy and value networks for MAgent2 battles, using self-play and a clipped policy objective to balance exploration with stable updates.

Self-play / Policy optimizationGitHub ↗
Language models06

English–Vietnamese
machine translation

Applied Direct Preference Optimization to EnvitT5 and mBART50, constructing preference pairs and evaluating translation quality through BLEU-based experiments.

DPO / Transformers / EvaluationGitHub ↗

Research & industry

Experience.

MBZUAI

Graduate Researcher

Conducting research on efficient machine learning systems and high-performance computing.

BioTuring

Software Engineer, Algorithms

Built lossless compression and streaming-quantile algorithms across four NVIDIA H100 GPUs for sparse biological data containing 20 million cells × 45,000 genes.

MOREH

GPU Trainee

Optimized model training and inference on NVIDIA and AMD accelerators using GPU profiling, kernel optimization, and multi-GPU parallelism.

Teko

Data Scientist Intern

Fine-tuned DeepSeek and LLaMA variants and built retrieval-augmented generation workflows with domain data and a Pinecone vector database.

Wisdom Brain Solution

AI Research Intern

Applied Direct Preference Optimization to English–Vietnamese translation, improving BLEU by 23 points and back-translation BLEU by 16 points.

Tools of the trade

Technical toolkit.

Programming & systems

Python · C++ · CUDA C/C++ · HIP · SQL
CUDA · ROCm · NCCL/RCCL · Multi-GPU computing

Machine learning & research

PyTorch · Hugging Face · Transformers · DPO
Tensor decomposition · Streaming algorithms · Statistical testing

Development & profiling

Linux · Git · Docker · NumPy · Pandas · scikit-learn
Nsight Systems/Compute · PyTorch Profiler · rocprof

Along the way

Selected distinctions.

  • ICPC VietnamNational Second Prize, 2023 & 2024 · Regional Third Prize, 2024
  • Vietnam National Olympiad in InformaticsSecond Prize, 2020 & 2021
  • Vietnam Student Informatics OlympiadSuper Cup Division · Third Prize, 2025
  • Academic scholarshipsVietcombank, 2025 · Odon Vallet, 2021
    University merit scholarships, 2023–2025

Let’s connect

Good research starts
with a conversation.

Interested in efficient AI, GPU systems, or a research collaboration? I’d be happy to hear from you.

Get in touch nqd250104@gmail.com