Computer Science PhD student at Vanderbilt University. Adversarial machine learning researcher.
Curious about everything. Skillful at team-working and collaborating. Creative and persistent.
Keen on self-learning. Passionate about sharing knowledge. CTF enthusiast.
Improving Routing in Sparse Mixture of Experts with Graph of Tokens
Tam Minh Nguyen, Ngoc N. Tran, Khai Nguyen, Richard Baraniuk
Preprint
|
Code
We stablize Mixture-of-Expert assignment fluctuations during training using information from token computation graphs.
Under review.
Beyond Losses Reweighting: Empowering Multi-Task Learning via the Generalization Perspective
Hoang Phan, Lam Tran, Quyen Tran, Ngoc N. Tran, Tuan Truong, Qi Lei, Nhat Ho, Dinh Phung, Trung Le
Preprint
|
Code
We leverage multiple loss formulations to enhance model generalization ability in multi-task learning.
Under review.
Generalization Bounds for Robust Contrastive Learning: From Theory to Practice
Ngoc N. Tran*, Tung Lam Tran*, Anh Tuan Bui, Tung Pham, Toan Tran, Dinh Phung, Trung Le
Preprint
|
Code
We identify which components in the unsupervised training process can improve the robust supervised loss.
Under review.
PBP: Post-training Backdoor Purification for Malware Classifiers
Dung Thuy Nguyen, Ngoc N. Tran, Taylor T. Johnson, Kevin Leach
Preprint
|
Proceedings
|
Code
We leverage batch statistics during finetuning to purify backdoored malware classifiers with limited data.
NDSS 2025.
Anti-DreamBooth: Protecting Users From Personalized Text-to-Image Synthesis
Thanh Van Le*, Hao Phung*, Thuan Hoang Nguyen*, Quan Dao*, Ngoc N. Tran, Anh Tran
Preprint
|
Proceedings
|
Code
|
Website
We propose adding imperceptible patterns to public images to void them from fine-tuning diffusion models.
ICCV 2023.
Sharpness & Shift-Aware Self-Supervised Learning
Ngoc N. Tran, Son Duong, Hoang Phan, Tung Pham, Dinh Phung, Trung Le
Preprint
|
Code
We enhance SimCLR by minimizing loss sharpness and improving the practical positive-pair distribution.
arXiv 2023.
Stochastic Multiple Target Sampling Gradient Descent
Hoang Viet Phan, Ngoc N. Tran, Trung Le, Toan Tran, Nhat Ho, Dinh Phung
Preprint
|
Proceedings
|
Code
|
Poster
We propose MT-SGD, allowing us to sample from the joint high-likelihood of multiple target distributions.
NeurIPS 2022.
Multiple Perturbation Attack: Attack Pixelwise Under Mixed Lp-norms For Better Adversarial Performance
Ngoc N. Tran, Anh Bui, Dinh Phung, Trung Le
Preprint
|
Code
|
Poster
We propose combining perturbations under different norms for a more effective yet imperceptible attack.
arXiv 2022.
ReINTEL Challenge 2020: A Comparative Study of Hybrid Deep Neural Network for Reliable
Intelligence Identification on Vietnamese SNSs
Hoang Viet Trinh, Tung Tien Bui, Tam Minh Nguyen, Huy Quang Dao, Quang Huu Pham, Ngoc N. Tran, Ta Minh Thanh
Preprint
|
Proceedings
We propose a multimodal model leveraging both tabular metadata and text content for fake news detection.
VLSP 2020.
Efficient Low-Latency Dynamic Licensing for Deep Neural Network Deployment on Edge Devices
Toan Pham Van, Ngoc N. Tran, Hoang Pham Minh, Tam Nguyen Minh, Thanh Ta Minh
Preprint
|
Proceedings
|
Slides
We proposed an efficient system for model version management and licensing for edge-device deployment.
CIIS 2020.
From Universal Language Model to Downstream Task: Improving RoBERTa-Based Vietnamese Hate Speech Detection
Quang Huu Pham, Viet Anh Nguyen, Linh Bao Doan, Ngoc N. Tran, Thanh Ta Minh
Preprint
|
Proceedings
|
Slides
We fine-tuned PhoBERT for Hate Speech Detection using various training techniques.
Best Paper Award, IEEE KSE 2020.
Interpreting the Latent Space of Generative Adversarial Networks using Supervised Learning
Toan Pham Van, Tam Minh Nguyen, Ngoc N. Tran, Hoai Viet Nguyen, Linh Bao Doan, Huy Quang Dao, Thanh Ta Minh
Preprint
|
Proceedings
|
Slides
|
Presentation
We used orthogonality constraints as regularization to enforce independent and interpretable embeddings.
ACOMP 2020.
Efficient Palm-Line Segmentation with U-Net Context Fusion Module
Toan Pham Van, Son Trung Nguyen, Linh Bao Doan, Ngoc N. Tran, Ta Minh Thanh
Preprint
|
Proceedings
|
Slides
We built a U-Net model with our custom Context Fusion Module to extract palm lines from hand images.
ACOMP 2020.
Deep Learning Approach for Singer Voice Classification of Vietnamese Popular Music
Toan Pham Van, Ngoc Tran Ngo Quang, Ta Minh Thanh
Preprint
|
Proceedings
|
Slides
We applied vocal segment detection, extraction, and a classifier to the singer voice classification problem.
SoICT 2019.
Deep Neural Networks Based Invisible Steganography for Audio-into-Image Algorithm
Quang Pham Huu, Thoi Hoang Dinh, Ngoc N. Tran, Toan Pham Van, Thanh Ta Minh
Preprint
|
Proceedings
|
Slides
We proposed a new approach of employing deep learning in hide secret audio in a cover image.
IEEE GCCE 2019.
This project is a reimplementation of
Chen et. al black-box decision-based
attack on deep learning models. Wrote a
post
explaining the paper. Created a related AI-centered CTF challenge.
This project implements an experimental deep model for extracting the vocal track from a
master track. Incorporated into our SoICT 2019 paper.
This project is composed of deep learning models that are trained to embed a
secret voice audio onto a cover image, or recover one from an image carrier,
respectively. Results from our IEEE GCCE 2019 paper.
This project implements a secure method of matching two people's lists of files and
finding the ones they have in common without exposing any details about the rest, using
the Paillier Cryptosystem and a secure hash function.
This project models the problem of assigning date/time, location, and board of professor
for each senior student as a massive quasilinear programming problem, then solve it
using AMPL/CPLEX.
This project does natural deductions on given puzzles and bruteforce all solutions when
needed. The user can try different initial states of the board and get the corresponding
possible solutions, aiding the research.