Meng Xu

Meng Xu

Assistant Professor | Department of Computer Science and Technology | Kean University

Building a Trustworthy Deep Learning Model for Urban-Scene Image Segmentation: Robustness and Uncertainty Analysis

Funding

CAHSI-Google Institutional Research program

Period

9/1/2024 to 8/31/2025

Role

Principal Investigator

Overview

This project examined trustworthy deep learning for urban-scene image segmentation, with attention to robustness and uncertainty analysis. It evaluated segmentation models’ robustness and generalization under adversarial perturbations using uncertainty and segmentation performance. The project also focused on ContraDiff, a contrastive learning diffusion model for trustworthy urban-scene image segmentation.

Team

  • Principal Investigator: Dr. Meng Xu
  • Co-Principal Investigator: Dr. Yuyin Zhou (University of California, Santa Cruz)
  • Undergraduate Student Researchers:
    • Cesar Marte

Acknowledgment

This project was funded by the CAHSI-Google Institutional Research program.