Tian Zhihua - Healthcare AI, Deep Learning, Action Recognition

Education

Master's in Operations Research and Control Theory
Nankai University (Sep 2023 - Present)
Key Courses: Operations Research & Optimization, Stochastic Processes, Computer Vision, Machine Learning, Deep Neural Networks
Bachelor's in Computer Science and Technology
Hangzhou Dianzi University (Sep 2017 - Jun 2021)

AI Diagnosis for Parkinson's Disease Using Video Analysis
Collaborated with healthcare professionals to develop a novel AI system for the intelligent diagnosis of Parkinson's disease based on video gait analysis. Leveraging single-camera videos of patients, conducted keypoint detection using dwpose to extract critical gait features. Utilized Graph Convolutional Networks (GCN) and an attention-based spatiotemporal mechanism to classify the gait data and evaluate the severity of the disease.

This work also included building a remote video diagnostic platform to facilitate patient monitoring and assessment in real-time. The AI model was successfully deployed on a cloud-based system, allowing medical professionals to conduct preliminary assessments remotely, thereby improving accessibility for patients in rural areas. This innovative approach received positive recognition from healthcare experts and demonstrated potential for scaling up in broader neurological disorder diagnostics.

Huawei Ascend AI Innovation Competition 2023
Tianjin Division, Application Track - Bronze Award for effective application of AI in healthcare, specifically in intelligent diagnostic solutions.
Tianchi Modelscope-Sora Challenge
Achieved 3rd Place in this highly competitive event, focused on video generation and multimodal learning. Developed innovative techniques for generating complex video scenarios and integrating multiple modalities to enhance model performance.

Research Engineer
Huawei Hangzhou Research Institute, AI Application Development Dept. (Apr 2023 - Jul 2023)
- Led research on high-resolution multi-angle portrait data generation using Unreal Engine, significantly reducing labeling costs for machine learning datasets.
- Enhanced facial recognition models in security scenarios using diffusion models to generate synthetic data and improve model accuracy.
- Performed model quantization to optimize visual models for 8-bit deployment, addressing accuracy loss while maintaining computational efficiency.
Algorithm Engineer
Hangzhou WoYing Network Technology Co., WoTu LAB (Apr 2021 - Apr 2022)
- Developed a facial attribute analysis model to address the lack of diverse datasets, focusing on Asian facial attributes and integrating them into a production pipeline.
- Created an AI image quality evaluation model that automated image selection, integrated as a key feature in production systems.
- Implemented AI slimming and enhancement algorithms using body detection and pose estimation technologies, contributing to feature expansion in consumer applications.