GRGBanking
Date:
He interned at the Research Institute of GRGBanking Equipment Co., Ltd., serving as an intern in the Data Elements Group and participating in project design and development from March 2025.
Project Description: Developed a privacy computing platform for data spaces, enabling each node to train a global model through federated learning by sharing parameters rather than raw data.
Privacy Computing × Healthcare: Addressed medical data silos by designing a federated learning platform for biomedical imaging. Compared federated learning, secure multi-party computation, and trusted execution environments. Conducted in-depth research on mainstream algorithms (Densenet201, ViT, ResNet50) for medical image classification. Integrated open-source tools (e.g., CVAT) for annotation and segmentation. Implemented cross-modal image-text retrieval and report generation via MedImageInsight’s dual-tower architecture, followed by user testing.
Data Value Evaluation: Designed a three-stage data pricing product, integrating national data quality standards and user cost-value analysis. Employed Principal Component Analysis (PCA) and Game Theory Approach for quotation formulation. Leveraged the Qirana-SQL pricing system, improved Shapley value, Dealer framework, and zero-payment models, combined with Trusted Execution Environment (TEE), federated learning, secure multi-party computation, and differential privacy. Developed valuation schemes for one-to-many/many-to-one/many-to-many scenarios between data owners and purchasers. Proposed a bargaining iteration scheme inspired by stock market transaction mechanisms.