Long-Giang Vu

I am a Computer Science Master’s student at UC San Diego. My intended research focus is on motion planning, specifically bridging foundation models with control theory for safety-guaranteed autonomous behavior. At UCSD, I work with Professor Ravi Ramamoorthi on 3D reconstruction and novel view synthesis. Previously, I conducted undergraduate research under the supervision of Professor Khoat Than on diffusion models for image generation and ML theory regarding the generalization capabilities of deep neural networks. I bring industry experience from an ADAS internship at Woven by Toyota and AI roles at TopCV, Viettel Group and Techcom Securities.

news

Mar 14, 2026	I’m thrilled to be returning to Woven by Toyota this April as a Machine Learning Engineer.
Jun 30, 2025	I am very glad to start a new role as a software engineer intern at Woven by Toyota this summer.
Mar 26, 2025	Paper Gentle Local Robustness implies Generalization is accepted at Springer Nature Machine Learning journal.
Aug 03, 2024	This September, I will start pursuing my Masters of Science in Computer Science and Engineering at University of California, San Diego (UCSD).
Jul 06, 2024	I am very honored to be awarded a VinGroup Scholarship to pursue my Master’s degree in the US.
Dec 09, 2023	Paper On Inference Stability for Diffusion Models is accepted at AAAI2024.

selected publications

AAAI

On Inference Stability for Diffusion Models

Giang Vu^*, Viet Nguyen^*, Tung Nguyen Thanh, and 2 more authors

AAAI Conference on Artificial Intelligence, 2024

Oral Abs PDF

Oral Presentation

Denoising Probabilistic Models (DPMs) represent an emerging domain of generative models that excel in generating di- verse and high-quality images. However, most current training methods for DPMs often neglect the correlation between timesteps, limiting the model’s performance in generating images effectively. Notably, we theoretically point out that this issue can be caused by the cumulative estimation gap between the predicted and the actual trajectory. To minimize that gap, we propose a novel sequence-aware loss that aims to reduce the estimation gap to enhance the sampling quality. Furthermore, we theoretically show that our proposed loss function is a tighter upper bound of the estimation loss in comparison with the conventional loss in DPMs. Experimental results on several benchmark datasets including CIFAR10, CelebA, and CelebA-HQ consistently show a remarkable improvement of our proposed method regarding the image generalization quality measured by FID and Inception Score compared to several DPM baselines. Our code and pre-trained checkpoints are available at https://github.com/VinAIResearch/SA-DPM.
SNML

Gentle Local Robustness implies Generalization

Khoat Than, Dat Phan, and Giang Vu

Springer Nature Machine Learning, 2025

Abs PDF

Robustness and generalization ability of machine learning models are of utmost importance in various application domains. There is a wide interest in efficient ways to analyze those properties. One important direction is to analyze connection between those two properties. Prior theories suggest that a robust learning algorithm can produce trained models with a high generalization ability. However, we show in this work that the existing error bounds are vacuous for the Bayes optimal classifier which is the best among all measurable classifiers for a classification problem with overlapping classes. Those bounds cannot converge to the true error of this ideal classifier. This is undesirable, surprizing, and never known before. We then present a class of novel bounds, which are model-dependent and provably tighter than the existing robustness-based ones. Unlike prior ones, our bounds are guaranteed to converge to the true error of the best classifier, as the number of samples increases. We further provide an extensive experiment and find that two of our bounds are often non-vacuous for a large class of deep neural networks, pretrained from ImageNet.