I am currently a first-year Ph.D. student in Computer Science at the Ohio State University, supervised by Prof. Wei-Lun Chao. I am generally interested in machine learning and computer vision for open-world machine learning with imperfect data.
I obtained my bachelor degree in Computer Engineering at University of Information Technology, Vietnam National University Ho Chi Minh City in late 2020. From 2022 to 2024, I was a research assistant at VinUni-Illinois Smart Health Center (VISHC), College of Engineering and Computer Science, VinUni, and AI research resident at FPT Software AI Residency, mentored by Prof Dung D. Le (VinUniversity).
My contact email: nguyen_dot_2959_at_osu_dot_edu
My research focuses on meta-learning (e.g., zero-shot/few-shot learning), uncertainty estimation (e.g., out-of-distribution detection, distribution-shift uncertainty), and domain adaptation (e.g., continual learning, domain distillation) in the context of general machine learning and computer vision problems. I am particularly interested in developing methods for learning with imperfect data (e.g., limited, noisy, or imbalanced datasets) under minimal human supervision (e.g., semi-supervised learning), while enabling effective extrapolation to unseen domains.
Additionally, I am also interested in eploring the intersection of black-box optimization and uncertainty estimation to advance lifelong and open-world learning systems.
Details of my research interests are discussed here.
Revisiting Semi-Supervised Learning in the Era of Foundation Models
Zheda Mai*, Ping Zhang*, Quang-Huy Nguyen, Wei-Lun Chao
Preprint, 2025
We introduce a simple yet effective Semi-supervised learning (SSL) approach by incorporating Vision Foundation Models (VFMs) and Parameter-Efficient Fine-Tuning (PEFT) in an ensemble manner for generating pseudo-labels for SSL self-training, resulted in a simpler yet more reliable self-training pipeline that effectively leverges both labeled and unlabled data.
Lessons and Insights from a Unifying Study of Parameter-Efficient Fine-Tuning (PEFT) in Visual Recognition
Zheda Mai, Ping Zhang, Cheng-Hao Tu, Hong-You Chen, Quang-Huy Nguyen, Li Zhang, Wei-Lun Chao
CVPR, 2025 Highlight (2.98%).
Our work proposes a systematic study for parameter-efficient fine-tuning (PEFT) methods, unveiling their performance in both few-shot/many-shot and distribution shift tasks toward suitable practical applications via a set of empirical evaluations and recommendations.
Improving Pareto Set Learning for Expensive Multi-objective Optimization via Stein Variational Hypernetworks
Minh-Duc Nguyen, Phuong Mai Dinh, Quang-Huy Nguyen, Long P. Hoang, Dung D. Le
AAAI, 2025
Our work explores Expensive Multi-Objective Optimization by introducing the Stein Variational Hypernetwork for Pareto Set Learning to overcome fragmented and uncertain regions in surrogate models while still maintain the diversity of the learned solutions, offering promising results for expensive multi-objective optimization problems.
Detecting Out-of-Distribution Objects through Class-Conditioned Inpainting
Quang-Huy Nguyen*, Jin Zhou*, Zhenzhen Liu*, Huyen Bui, Kilian Q. Weinberger, Wei-Lun Chao, Dung D. Le
Preprint, 2025
Our work addresses OOD Object Detection by exploiting the inconsistency between generative and discriminative model outputs. Using off-the-shelf generative model as an auxiliary to object detector and introducing a triplet similarity metric that captures both semantic and visual differences, our method effectively identifies OOD objects without any re-training.
Controllable Expensive Multi-objective Learning with Warm-starting Bayesian Optimization
Quang-Huy Nguyen*, Long P. Hoang*, Hoang V. Vu, Dung D. Le
Preprint, 2024
Our work explores Multi-objective Black-box Optimization with Pareto Front Learning, aligning trade-off preferences with corresponding optimal solutions between conflicting objectives by leveraging warm-starting Bayesian Optimization to adequately obtain a good approximation of the front first and re-initialize the Pareto Set Model during the opimization steps to stabilize the PSL.
Enhancing Few-shot Image Classification with Cosine Transformer
Quang-Huy Nguyen, Cuong Q. Nguyen, Dung D. Le, Hieu H. Pham
IEEE Access, 2023
We explore Few-shot Image Classification by proposing a new cross-attention mechanism based on cosine similarity, without using softmax, to further emphasizes the correlation between labeled supports and unlabeled query representations, thus enhancing ViT-based few-shot algorithms across various settings and scenarios compare to convention attention mechanism.
Apr, 2025: Our paper Lessons and Insights from a Unifying Study of Parameter-Efficient Fine-Tuning (PEFT) in Visual Recognition is accepted at CVPR 2025 as Highlight.
Dec, 2024: Our paper Improving Pareto Set Learning for Expensive Multi-objective Optimization via Stein Variational Hypernetworks is accepted at AAAI 2025.
Aug, 2024: I become PhD student in Computer Science and Enginering at the Ohio State University, advised by Prof. Wei-Lun (Harry) Chao.
Aug, 2023: I become AI Research Resident at FPTSoftware AI Center, Ho Chi Minh City.
Jul, 2023: Our paper Few-shot Cosine Transformer is accepted at IEEE Access.
Feb, 2023: I become Research Assistant at College of Engineering and Computer Science, VinUniversity.
Jan, 2022: I become Research Assistant at VinUni-Illinois Smart Health Center (VISHC), VinUniversity.
Dec, 2020: I obtain a Bachelor degree in Computer Engineering at University of Information Technology, VNU-HCM.