Zihan(Z) Dong

AN INNOVATIVE ENGINEER WITH A PASSION FOR MAKING A POSITIVE IMPACT ON THE WORLD.

About me 🤗

I am a Workstation Engineer at Lenovo and a graduate of the Georgia Institute of Technology’s School of Computer Science. My research and engineering focus lies at the intersection of generative AI, proactive agents, and human–computer interaction. I design and build memory-augmented, multimodal agent systems to support workflows in productivity, education, and next-generation “AI PC” experiences. Driven by a belief in accessible, robust open-source AI, I’m committed to bridging cutting-edge research with real-world impact.

Education 📚

Program & Projects 🛠️ 🤖️

Running Projects

See All Projects →

Publications 📑 (See All)

Avoid Catastrophic Forgetting with Rank-1 Fisher from Diffusion Models

Z Wang, A Gupta, Zihan Dong, CJ MacLellan

arXiv preprint • 2025

We propose Rank-1 Fisher, a low-rank, diffusion-based approximation of the Fisher Information Matrix that reduces catastrophic forgetting while preserving performance. Improves continual learning stability and reduces memory overhead.

Towards General Computer Control with Hierarchical Agents and Multi-Level Action Spaces

Zihan Dong, X Fan, Z Tang, Y Li

arXiv preprint • 2025

Introduces a hierarchical agent framework for multi-level GUI and keyboard/mouse control. Enables multi-step planning, adjustable abstraction levels, and general-purpose computer interaction.

FNSPID: A Comprehensive Financial News Dataset in Time Series

Zihan Dong, Xinyu Fan, Zhiyuan Peng (Under Review)

KDD 2024 Applied Data Science Track

We introduce FNSPID: 29.7M stock prices and 15.7M time-aligned financial news for 4,775 S&P500 companies (1999–2023). Experiments show scale and quality improve prediction accuracy; adding sentiment modestly boosts transformer-based models. Code and data: github.com/Zdong104/FNSPID.

Enhancing Bloodstain Analysis Through AI-Based Segmentation

Zihan Dong, Zhengdong Zhang

RelKD 2023 Workshop at KDD 2023

We evaluate pre-trained and fine-tuned SAM for bloodstain image segmentation. Fine-tuned SAM improved accuracy by 2.2% and speed by 4.7%. Code: Bloodstain_Analysis_Ai_Tool.

Students' Perceptions and Preferences of Generative Artificial Intelligence Feedback for Programming

Zihan Dong, Zhengdong Zhang, Yang Shi, Noboru Matsuda, Thomas Price, Dongkuan Xu

AAAI 2024

We explore ChatGPT for formative feedback on CS1 Java assignments via surveys from 102 students, comparing prompts with/without code and summarizing improvements to make AI-generated feedback more useful.

[Abstract] Exploring the Augmented Large Language Model with Mathematical tools in Personalized and Efficient Education

with Dongkuan Xu · ICAIBD 2023

We augment ChatGPT with math performance assessments to personalize learning and study bias, proposing a framework for efficient, tailored education.

[Abstract] Random Matrix Theory (RMT) to quantify scattering behavior in Lung mimicking phantoms 🫁

Zihan Dong, Azadeh D. Cole, Henry Ware, Marie Muller · ASA 2023

RMT parameters from SVD of inter-element matrices correlate with histology; PEGDA phantom experiments indicate multiple scattering increases with pore size.

Research Experience 🔍

MoLab 💪

As an undergraduate, I processed musculoskeletal data with MATLAB and OpenSim to investigate upper limb/hand function, built datasets and algorithms for bio-electric signal personalization, and prepared data for publication.

NC State and UNC Joint Department of Biomedical Engineering 🐷🦵

Processed porcine hindlimb MRIs with 3D Slicer/ITK-SNAP, performed epiphysis segmentation, built 3D models (STL), measured notch/condyle widths, and ran ShapeWorks for SSM to compare intact vs ACL-injured knees.

Awards and Honor