CV

Doctoral Researcher

CVI2, SnT, University of Luxembourg

• Robust and Generalizable Deepfake Detection
• Mentoring Master's Students
• Publication Venues: CVPR, ICCV

Technical Lead & Product Owner

VinAI Research

• Traffic Sign/Light Detection: ~300 classes, AGX Drive, CV25 Ambarella, Day-Night time, Occlusion, Extreme Lighting Conditions, Tiny Objects, Imbalanced Data
• 2D Pose Estimation & Action Recognition: Distribution-Aware Keypoint Estimation, COCO, Violence Detection, 3D Keypoints, Zero-shot Learning

AI Research Engineer

Sun Asterisk Japan

• Liquors Recommendation: Distribution-Aware Content-Based Recommendation Systems, Data Crawling & Pre-processing
• Publication Venues: Applied Intelligence (ISI-Q1), ICCCI (B), JSTIC (Domestic)

Technical Lead

Sun Asterisk Vietnam

• End-to-End Face Recognition, Facial Matting, Wrist Detection, and Deep Learning Model Optimization
• Online Learning for ML System, Image Retrieval, Real-time Inference, Lightweight Segmentation
• ElasticSearch, Django backend APIs, Deployment, NginX
• Publication Venues: ICCSA (B), Computer Science (Q3)

Software Engineer

Framgia Inc.

• Backend: Ruby on Rails, Python - Django, RESTful APIs
• Frontend: JavaScript - ReactJS, JQuery, HTML, CSS, Bootstrap
• Database: Elasticsearch, MongoDB, MySQL, Firebase
• Management Tools: Github, Jira, Trello
• Deployment: NginX
• Project Management: Agile

Intern

Framgia Inc.

• Completed intensive training in Ruby on Rails, JavaScript, HTML, CSS, and contributed to internal projects under supervision

Doctoral

University of Luxembourg

Computer Science

Master

VietNam National University - VNU

Computer Science

Bachelor

Military Technical Academy

Software Engineering

High School

Nguyen Hue High School for the Gifted

Specialized Class in Programming

Best Paper Award, LAA-Net

Faculty of Science, Technology, and Medicine - Uni.lu

LAA-Net received the best paper award at the DPCSCE PhD Research Day, an annual event for 2nd-year PhD students.

1st Runner-Up Sun* Hackathon

Sun Asterisk

The competition last for 36 hours. Our mission was to play role as a start-up, from the idea brain-storm, project design, implementation to presentation. Finally, we stand 2nd position amongst 20 teams.

GameLoft scholarship for excellent students

GameLoft

Hot trainee of the year

Sun Asterisk

Best trainee of the month

Sun Asterisk

Vulnerability-Aware Spatio-Temporal Learning for Generalizable Deepfake Video Detection

ICCV 2025

Fakeformer: Efficient vulnerability-driven transformers for generalisable deepfake detection

arXiv preprint arXiv:2410.21964

LAA-Net: Localized Artifact Attention Network for Quality-Agnostic and Genelizable Deepfake Detection

CVPR 2024

HYBRID END-TO-END APPROACH INTEGRATING ONLINE LEARNING WITH FACE-IDENTIFICATION SYSTEM

Computer Science

Multi-label deepfake classification

MMSP 2023

A Novel Approach to End-to-End Facial Recognition Framework with Virtual Search Engine ElasticSearch

ICCSA 2021

Solving distribution problems in content-based recommendation system with gaussian mixture model

Applied Intelligence

Robust content-based recommendation distribution system with gaussian mixture model

ICCCI 2020

Deepfake Detection

In this project, I focused on developing advanced methods to detect high-quality and unseen deepfakes, including those not encountered during training. Our approaches not only demonstrated superior performance on multiple datasets compared to state-of-the-art methods but also showed greater robustness to unseen common perturbations (e.g., Gaussian noise). Notably, our models require minimal computational overhead, underscoring their practical usefulness. This work has resulted in three publications, including two accepted at top-tier conferences in Computer Vision.

• Generalizable to Unseen Manipulations
• Robustness to Unseen Perturbations, e.g., Gaussian Noise, etc
• Minimal Computational Overead
• Real-time Inference

Violence Detection

Designing a model that receives a video as input and predicts the probability of violent actions.

• Levergaging Object Keypoint Similarity (OKS) as a Trigger
• Spatio-Temporal Modelling
• Zero-shot Learning

Traffic Sign/Light Detection

One of the most challenging projects I have worked on involved building a detection model that is robust to varying conditions such as weather, daytime, nighttime, and extreme lighting. The model also needed to be computationally optimal for deployment on AGX Driver or ARM64 platforms. In total, it had to handle around 300 classes of traffic signs and lights.

• ~300 Fine-grained Classes
• Day-Night time, Occlusion, Extreme Lighting Conditions, Tiny Objects, Imbalanced Data
• ONNX, TensorRT, AGX Driver, ARM64

Liquors Recommendation

In this project, I worked directly with a customer in Japan to develop a recommendation algorithm that suggests similar liquors based on taste coefficients. The algorithm ultimately outperformed a well-known sake recommendation website (https://sakenowa.com/). This work also resulted in two publications in prestigious venues.

• Distribution-Aware Content-Based Recommendation Systems
• Data Crawling & Pre-processing
• Improved Recommendation Results as compared to Sakenowa (https://sakenowa.com/)

Face Recognition

I developed a face recognition system that requires only CPUs for training on a small dataset, yet achieves accuracy and speed comparable to state-of-the-art methods. Moreover, the system supports online learning. This work has resulted in two publications.

• Online Learning
• Image Retrieval with Elasticsearch
• Training: CPUs only and not require Large-scale Datasets
• Real-time Inference
• Django backend APIs
• Deployment: NginX

Insight Data Science

In this project, I focused on the front-end, using ReactJS to display content returned from back-end APIs.

• Front-End
• ReactJS

JAMJA

For this project, I worked on the back-end, developing APIs to serve the front-end. My main task was integrating Elasticsearch to build a multi-level search bar with diverse criteria such as location, proximity, popularity, and brand.

• Backend with Django
• Elasticsearch, MongoDB, Firebase

Collatotte

This was the first project I joined as a Web Developer after completing my bachelor’s degree. I was responsible for building the admin site, enabling website administrators to manage products through a reliable and user-friendly interface.

• Ruby On Rails
• JavaScript - ReactJS, JQuery, HTML, CSS, Bootstrap
• MySQL