Edge AI & Privacy
Model partitioning, distributed inference, and privacy-aware visual computing.
Academic Portfolio
Zihao Ding
PhD Student, ECE, Rutgers University
I am a PhD student in Electrical and Computer Engineering at Rutgers University, working on AR/VR systems and edge ML under the supervision of Prof. Yao Liu. My research focuses on running computation-intensive vision and AI models on resource-constrained personal devices through model partitioning and systems optimization.
I also previously integrated real-world sensing and infrastructure systems, including the WIM (Weigh-in-Motion) deployment on the BQE. Broadly, I work at the intersection of systems, multimedia, and hardware-aware machine learning.
RESEARCH SNAPSHOT
Themes
My work studies how personal and immersive devices can run demanding AI and visual computing workloads under privacy, latency, and resource constraints.
AR/VR & Egocentric Systems
Systems support for immersive navigation, egocentric sensing, and long-horizon video understanding.
Multimedia Systems
Measurement, datasets, and replay tools for visual computing pipelines and interactive media workloads.
LLM Agent System Performance
Characterizing latency, cost, and network behavior in tool-augmented agent workflows.
Selected Publications
Paper Catalog
Selected work across edge AI, AR/VR systems, multimedia systems, and agent infrastructure.
Egocentric Daily Video Question Answering with Token-Efficient Storyboard Retrieval
Proceedings of the 36th Workshop on Network and Operating System Support for Digital Audio and Video, NOSSDAV 2026
A token-efficient retrieval pipeline for long egocentric videos using compact storyboard representations to preserve visual context for question answering while reducing token cost.
A Distributed Framework for Privacy-Enhanced Vision Transformers on the Edge
ACM/IEEE SEC 2025
A distributed Vision Transformer framework that partitions attention computation across multiple servers to improve visual privacy while preserving near-baseline model accuracy.
EyeNavGS: A 6-DoF Navigation Dataset and Record-n-Replay Software for Real-World 3DGS Scenes in VR
ACM Multimedia 2025
A real-world 6-DoF VR navigation dataset for 3D Gaussian Splatting scenes, with record-and-replay software for evaluating rendering and VR system performance.
Network and Systems Performance Characterization of MCP-Enabled LLM Agents
arXiv preprint arXiv:2511.07426
A systems-oriented characterization of MCP-enabled LLM agents, focusing on network behavior, latency, cost, and overheads in tool-augmented agent workflows.
Recent Updates
News
- NOSSDAV paper on token-efficient egocentric video QA.
- Privacy-enhanced Vision Transformer framework accepted to ACM/IEEE SEC.
- EyeNavGS published at ACM Multimedia.
- MCP-enabled LLM agent performance preprint released.
Links
Profiles & Projects
A few useful entry points for research profiles, code, and engineering work.