Jiemi Zhang

Didi Chuxing (China), Cornell University

Papers

Total Citations

H-Index

About

Jiemi Zhang is a researcher whose work lies at the intersection of computer vision, robotics, and unsupervised machine learning, with a particular focus on understanding and assisting with human daily activities. Zhang’s major contributions center on developing systems that can learn complex human behaviors without requiring labeled data. The most notable work, "Watch-n-Patch" (2017, 23 citations), introduces a completely unsupervised method for modeling composite human activities by learning both high-level co-occurrence and temporal relations between basic actions, addressing the challenge of large variation in everyday tasks. This foundational approach is complemented by the innovative "Watch-Bot" system (2016, 14 citations), which uses a simple RGB-D sensor to detect forgotten actions during an activity and proactively reminds the human by pointing a laser at the relevant object. This work demonstrates a practical application of unsupervised learning for assistive robotics, showing how a system can watch, infer, and intervene. Through these contributions, Zhang has advanced the field of activity understanding, enabling machines to autonomously learn and assist with the nuanced, unscripted nature of human life.