Hai Nguyen

Hi! I’m Hai. I’m an incoming ECE Ph.D. student at Cornell University. Before joining Cornell, I completed my undergraduate studies at Hanoi University of Science and Technology, majoring in Data Science and Artificial Intelligence. During that time, I also spent two wonderful years at Qualcomm AI Research as a Research Resident, where I was the youngest person ever to hold this position. At Qualcomm, I was fortunate to be advised by Prof. Nhat Ho and to work closely with Prof. Khai Nguyen.

Research: I am currently working on making AI systems safe, reliable, and interpretable. I am also very interested in efficient AI, designing algorithms that remain practical at scale. These are the some main research directions I have been pursuing:

1. Mechanistic Interpretability. I am currently working on methods to open the black box of AI systems.

2. Dataset-centric AI. I developed fast dataset similarity measures that help identify appropriate datasets and pretrained models for downstream tasks, supporting decisions in model selection, transfer, and data curation.

3. Optimal Transport. I previously worked on improving the scalability and fairness of Optimal Transport, with an emphasis on the Sliced Wasserstein family and its applications.

📬 I'm open to collaboration opportunities and would love to connect if our research interests align, feel free to reach out via email at hnn7@cornell.edu or namhai283287@gmail.com.

Selected Papers [Full List]

(*) denotes equal contribution

ICML

Lightspeed Geometric Dataset Distances via Projections

Hai Nguyen *, Khai Nguyen *, Tuan Pham, and Nhat Ho

International Conference on Machine Learning

Abstract PDF

We introduce sliced optimal transport dataset distance (s-OTDD), a model-agnostic, embedding-agnostic approach for dataset comparison that requires no training, is robust to variations in the number of classes, and can handle disjoint label sets. The core innovation is Moment Transform Projection (MTP), which maps a label, represented as a distribution over features, to a real number. Using MTP, we derive a data point projection that transforms datasets into one-dimensional distributions. The s-OTDD is defined as the expected Wasserstein distance between the projected distributions, with respect to random projection parameters. Leveraging the closed form solution of one-dimensional optimal transport, s-OTDD achieves (near-)linear computational complexity in the number of data points and feature dimensions and is independent of the number of classes. With its geometrically meaningful projection, s-OTDD strongly correlates with the optimal transport dataset distance while being more efficient than existing dataset discrepancy measures. Moreover, it correlates well with the performance gap in transfer learning and classification accuracy in data augmentation.
ICLR Spotlight

Towards Marginal Fairness Sliced Wasserstein Barycenter

Hai Nguyen *, Khai Nguyen *, and Nhat Ho

International Conference on Learning Representations
Spotlight Presentation [Top 3.2%]
Abstract PDF

The Sliced Wasserstein barycenter (SWB) is a widely acknowledged method for efficiently generalizing the averaging operation within probability measure spaces. However, achieving marginal fairness SWB, ensuring approximately equal distances from the barycenter to marginals, remains unexplored. The uniform weighted SWB is not necessarily the optimal choice to obtain the desired marginal fairness barycenter due to the heterogeneous structure of marginals and the non-optimality of the optimization. As the first attempt to tackle the problem, we define the marginal fairness sliced Wasserstein barycenter (MFSWB) as a constrained SWB problem. Due to the computational disadvantages of the formal definition, we propose two hyperparameter-free and computationally tractable surrogate MFSWB problems that implicitly minimize the distances to marginals and encourage marginal fairness at the same time. To further improve the efficiency, we perform slicing distribution selection and obtain the third surrogate definition by introducing a new slicing distribution that focuses more on marginally unfair projecting directions. We discuss the relationship of the three proposed problems and their relationship to sliced multi-marginal Wasserstein distance. Finally, we conduct experiments on finding 3D point-clouds averaging, color harmonization, and training of sliced Wasserstein autoencoder with class-fairness representation to show the favorable performance of the proposed surrogate MFSWB problems.