Yueming Zhu

-Improved river safety in Columbus through machine learning, computer vision, and digital twin models. -Developed a water level prediction model based on LSTM by historical data, capable of issuing possible alerts. -Implemented YOLO-v7 for human behavior recognition with live streaming from an on-site webcam. -Constructed a digital twin of the river and projected the camera-captured locations of people onto it. -Integrated the above components into a dashboard using the RTMP for webcam and Redis for variable sharing.

- Conducted advanced research in breast cancer diagnosis automation under the guidance of Dr. May Wang. - Investigated contrastive learning, data augmentation, and active learning methods for label-efficient problems. - Dived into contrastive learning and implemented SimCLR on different backbones to get the best performance. - Reached the fully supervised learning accuracy benchmark with only 20% of labeled datasets.

- “The Embedded Entry System by Palmprint Recognition” advised by Professor Dexing Zhong. - Deployed the whole entry system on the AIO-3399C board which can register and verify the user’s identity. - Trained the feature extractor by VGG16 in PyTorch, wrote the GUI by QT, and zipped the whole project up.

- Worked with Professor Qing Wang in the Department of Software Technology on the “Gesture Recognition Empowered by Ambient Light, Simple Photodiodes, and Embedded Artificial Intelligence” project. - Collected the gesture datasets using our self-designed and self-built circuit, then cleaned and organized them. - Built a gesture recognition model by TensorFlow and trained it, then converted it to binary by TensorFlow Lite. - Deployed the whole system in an embedded system and achieved real-time recognition of 8 gestures.