Quan Nguyen Minh

I am currently a second-year PhD student at the University of Florida, conducting research in trustworthy and secure AI, foundation models, and federated learning. Previously, I received my Bachelor degree in Biomedical Engineering from Hanoi University of Science and Technology and my Master degree in Biomedical Computing from Technical University of Munich (TUM).

Email  /  CV  /  Google Scholar  /  Github

I am passionate about trustworthy AI and scalable decentralized machine learning. Representative papers are highlighted.

We propose PromptMIA, a membership inference attack tailored to federated prompt-tuning, in which a malicious server can insert adversarially crafted prompts and monitors their updates during collaborative training to accurately determine whether a target data point is in a client's private dataset. We formalize this threat as a security game and empirically show that PromptMIA consistently attains high advantage in this game across diverse benchmark datasets. Our theoretical analysis further establishes a lower bound on the attack's advantage which explains and supports the consistently high advantage observed in our empirical results.

We derive theoretical lower bounds for the success rates of low-polynomial time MIAs that exploit vulnerabilities in fully connected or self-attention layers. We establish that even when data are protected by LDP, privacy risks persist, depending on the privacy budget. Practical evaluations on federated vision models confirm considerable privacy risks, revealing that the noise required to mitigate these attacks significantly degrades models' utility.

In this work, we introduce the Differentially Private ParameterShift Rule (Q-ShiftDP), the first privacy mechanism tailored to Quantum ML.

In this work, we present deep learning solutions for 3D intracardiac echocardiography segmentation and outline key challenges for real-time clinical deployment.

We propose FedDCT (Federated Divide and Collaborative Training), a novel FL architecture that allows for DCML with considerably lower client memory requirements over traditional FL while introducing no additional training cost on the central server. FedDCT enables a cluster of several clients to cooperatively train a large deep learning model, in contrast to standard FL, which demands that each client train the full-size neural network independently during each training round.

In this work, we propose ARPD, a single-stage anchor-free fully convolutional network to detect arbitrarily rotated people in fish-eye images. Our network uses keypoint estimation to find the center point of each object and regress the object's other properties directly. To capture the various orientation of people in fish-eye cameras, in addition to the center and size, ARPD also predicts the angle of each bounding box.

The aim of this work is to incorporate the de-warping of fisheye image using polar to Cartesian transformation to generate a panoramic view to aid in fall detection.

●   English (IELTS 8.5), German (A2)

●   2026 CISE Graduate Scholarship, University of Florida
●   2023 Scholarship - Modern Machine Learning: Foundations and Applications Program
●   2023 Scholarship - 9th Vietnam Summer School of Science
●   2022 First prize, Vietnam AI Day (VAIPE: Protective healthcare monitoring)
●   2021 Third prize, HUST Academic Research Contest
●   2020 Academic Incentive Scholarship, HUST
●   2019 Scholarship - FPT Center for Young Talents
●   2018 International Exchange Scholarship at Temasek Polytechnic

This repository contains the code and experiments for the paper FedDCT:Federated Divide and Collaborative Training, accepted to IEEE Transactions on Network and Service Management
Github

I worked in a team with other engineers from VHT to design face recogntion algorithms to participate in NIST FRVT 1:1 Verification Test. NIST provides independent evaluations of commercially available and prototype face recognition technologies. Our submission achieved respectable results.

I reimplemented the famous SORT tracking algorithm from scratch in Java using only OpenCV. Was a fun project to code and i applied this for tracking Vehicle on edge devices.
Github

I redesigned the classic Mobilenet-SSD network to reduce FLOPS and Params while keeping good accuracy. I pruned the backbone manually, modify the feature map size and anchor configs. The detector achieves 30 FPS on Intel CPU while keeping respectable accuracy.
Github

This collection of Python code provides implementations for various analytical tomographic image reconstruction algorithms. These algorithms play a crucial role in transforming projection data into detailed images, making them valuable tools in medical imaging, materials science, and more. Included are Fourier Gridding, Backproject Filter, Filtered Backprojection and Convolve Backproject
Github