Ph.D., Computational Science & Engineering

Georgia Institute of Technology · GPA 4.0/4.0

• Advised by Dr. J. David Frost and Dr. Max Mahdi Roozbahani.
• Ph.D. thesis: “Post-disaster Damage Assessment Based on Deep Learning with Surrogate Data.”

M.S., Computer Science

Georgia Institute of Technology · GPA 4.0/4.0

B.S., Software Engineering

University of Electronic Science and Technology of China (UESTC) · GPA 3.74/4.0

• Excellent Graduate of UESTC; China National Scholarship and study scholarships.

Postdoctoral Fellow

Georgia Tech · advised by Dr. David Frost & Dr. Mahdi Roozbahani

• Developing CHAIN (Cooperative Human–AI Nexus) for post-disaster image processing and analysis — integrating social signals, geospatial data, remote-sensing imagery, and infrastructure engineering information for comprehensive disaster assessment.
• Supporting area-of-interest prioritization, infrastructure-disruption assessment, exposure estimation, damage detection, disaster-region delineation, and recovery monitoring.
• Building a hybrid human–machine decision system that combines AI-based scoring with expert adjustment for standardized, explainable disaster-level determination.

Graduate Research Assistant

Georgia Tech · advised by Dr. David Frost & Dr. Mahdi Roozbahani

• Earthquake damage analysis from social-media images using self-supervised learning (MAE, MoCo, DINOv2) and large vision models (SAM, CLIP, Grounding DINO); built an end-to-end framework for domains with limited data and atypical visual features.
• Generated high-quality datasets for classification, detection, and semantic segmentation, with auto-annotation powered by generative AI (MLLMs).
• Analyzed Black Marble Nighttime Light (NTL) data for rapid disaster response and situational awareness.

Master Project

Georgia Tech · advised by Dr. Mahdi Roozbahani

• Developed multi-scale unsupervised representation learning (e.g., SimCLR, RotNet) optimized for low-data domains.

Undergraduate Researcher

University of Electronic Science and Technology of China

• Proposed an improved one-dimensional Logistic chaotic PRNG based on topological conjugation, improving the uniformity of the traditional generator.

Undergraduate Research Intern

Chengdu Vroad Software Technology

• Built an early fire-detection system for tunnels — 99% accuracy with SVM and YOLOv3 in C++.
• Implemented YOLOv3-based highway vehicle detection, 22% faster than SVM + HOG.

Lexical Knowledge Integration via Contrastive Learning

• Fine-tuned RoBERTa-base with the SimCSE framework on WordNet 3.0 to address low similarity between word and sentence embeddings.

Deep Reinforcement Learning for Tunable Agents

• Integrated tabular Q-learning with DNNs for a two-player zero-sum, high-dimensional pursuit–evasion game built on OpenAI Gym.