Reproducing Papers, One Model at a Time.
Building trustworthy AI for cybersecurity and healthcare — one reproduced paper at a time.
I'm a PhD-track researcher working at the intersection of AI for Cybersecurity, AI for Healthcare, Federated Learning, and Continual Learning, with a growing focus on efficient LLMs/VLMs and Edge AI/TinyML. My work centers on making machine learning systems that are private, robust, and deployable in resource-constrained, high-stakes environments — from intrusion detection on IoT devices to privacy-preserving diagnostic models across hospitals.
cat resnet-image-classification.md
Deep Residual Learning for Image Recognition (ResNet)
ResNet introduced residual connections that let networks train with far greater depth by reformulating layers as learning residual functions with reference to the layer inputs, addressing the degradation problem in very deep networks.
Deep Residual Learning for Image Recognition (ResNet)
ResNet introduced residual connections that let networks train with far greater depth by reformulating layers as learning residual functions with reference to the layer inputs, addressing the degradation problem in very deep networks.
An Image is Worth 16x16 Words (Vision Transformer)
ViT applies a pure Transformer architecture directly to sequences of image patches, showing that with sufficient pretraining data, CNN-specific inductive biases can be matched or exceeded by attention-based models.
Communication-Efficient Learning of Deep Networks from Decentralized Data (FedAvg)
FedAvg introduces a simple, communication-efficient algorithm for training a shared model across decentralized clients by averaging locally-computed updates, without centralizing raw data — the foundational algorithm of federated learning.
ls -la ./recent-activity/
ls -la ./projects/
Edge-NIDS: On-Device Intrusion Detection for IoT
A quantized CNN intrusion-detection model deployed to an ARM Cortex-M microcontroller for real-time IoT traffic monitoring without cloud connectivity.
FedDiagnose: Cross-Hospital Federated Diagnostic Modeling
A simulated multi-hospital federated learning platform for chest X-ray classification, keeping patient imaging data local to each institution.
ReproBench: Automated Paper Reproduction Scorecard
A tooling suite that runs reproduced model checkpoints against held-out benchmarks and auto-generates a comparison scorecard against reported paper results.
cat blog.log
Why I Reproduce Landmark Papers Before Doing Original Research
Reproduction is not busywork — it is the fastest way to internalize a method deeply enough to extend it responsibly. Here is the practice I follow for every paper I reproduce.
Reading ResNet Ten Years Later: What Still Holds Up
Residual connections are everywhere now, which makes it easy to forget how strange the degradation problem looked in 2015. A close re-read of the original paper.
Your Intrusion Detector Has the Same Blind Spot as an Image Classifier
FGSM was invented on image classifiers, but the same gradient-based blind spot shows up in tabular network-flow features. A walkthrough of adapting adversarial attacks to NIDS.
git log --oneline ./milestones
-
2024 — Present
PhD Track Researcher, AI Security & Health Lab
University of Technology, Ho Chi Minh City
Researching federated and continual learning methods for intrusion detection and clinical diagnostic models under data-scarcity and privacy constraints.
-
2023 — 2024
Research Assistant, Applied ML Group
University of Technology, Ho Chi Minh City
Built baseline pipelines for network traffic classification using deep learning; co-authored a workshop paper on adversarial robustness of NIDS models.
-
2022
B.Eng. Computer Science, Summa Cum Laude
University of Technology, Ho Chi Minh City
Thesis: "Lightweight Convolutional Architectures for On-Device Malware Detection." Graduated top of cohort.
-
2021
ML Engineering Intern
CyberShield Vietnam
Shipped a production anomaly-detection microservice for SOC alert triage, reducing analyst triage time by 35%.