Gia Huy Thai
Portrait
Undergraduate Researcher
Leader, Mixture of Multimodal Experts (MoME)

I am a final-year Honors Program undergraduate student in Information Technology at VNUHCM - University of Science. Separately, I lead research at the independent Mixture of Multimodal Experts (MoME) group, advised by Tuan Dang. My research interests include computer vision, medical image analysis, multimodal learning, and efficient expert-based neural architectures.

Curriculum Vitae
Email Google Scholar GitHub LinkedIn ORCID
Education
  • AI VIETNAM
    AI VIETNAM
    Extracurricular
    Research Assistant
    Feb. 2024 - present
  • VNUHCM - University of Science
    VNUHCM - University of Science
    Faculty of Information Technology | Honors Program
    Bachelor of Science student | GPA: 3.86/4.0
    Sep. 2023 - present
  • Hoang Le Kha High School for the Gifted, Tay Ninh
    Hoang Le Kha High School for the Gifted, Tay Ninh
    Specialized in Mathematics
    GPA (Grade 12): 9.5/10.0
    Sep. 2020 - May 2023
Honors & Awards
Selected Publications (view all )
SAGE: Shape-Adapting Gated Experts for Adaptive Histopathology Image Segmentation
SAGE: Shape-Adapting Gated Experts for Adaptive Histopathology Image Segmentation

Gia Huy Thai*, Hoang-Nguyen Vu*, Anh-Minh Phan, Quang-Thinh Ly, Tram Dinh, Thi-Ngoc-Truc Nguyen, Nhat Ho (* equal contribution)

CVPR 2026 Findings Track

An input-adaptive framework that reconfigures static backbones into dynamically routed expert architectures for state-of-the-art histopathology image segmentation.

[Project Page] [Paper] [Code]

SAGE: Shape-Adapting Gated Experts for Adaptive Histopathology Image Segmentation

Gia Huy Thai*, Hoang-Nguyen Vu*, Anh-Minh Phan, Quang-Thinh Ly, Tram Dinh, Thi-Ngoc-Truc Nguyen, Nhat Ho (* equal contribution)

CVPR 2026 Findings Track

An input-adaptive framework that reconfigures static backbones into dynamically routed expert architectures for state-of-the-art histopathology image segmentation.

[Project Page] [Paper] [Code]

ELDER: Equivariance-preserving Downsampling for Efficient Low-Resolution Image Classification
ELDER: Equivariance-preserving Downsampling for Efficient Low-Resolution Image Classification

Trong-Dai Phan, Gia Huy Thai, Minh-Huy Mai-Duc

ACMLC 2026

A systematic study of equivariance-preserving downsampling for efficient low-resolution CNN classification. Achieves 81.78% accuracy on CIFAR-10 with only 570K parameters through architectural variants that maintain spatial information.

[Paper]

ELDER: Equivariance-preserving Downsampling for Efficient Low-Resolution Image Classification

Trong-Dai Phan, Gia Huy Thai, Minh-Huy Mai-Duc

ACMLC 2026

A systematic study of equivariance-preserving downsampling for efficient low-resolution CNN classification. Achieves 81.78% accuracy on CIFAR-10 with only 570K parameters through architectural variants that maintain spatial information.

[Paper]

All publications