Towards Contactless Data-Model Matching

Data-model matching, typically achieved through direct contact, is critical to digital markets. However, when data and models belong to different owners, the direct contact-based form faces some security threats, including data security, privacy disclosure, and model reverse engineering attacks. A natural question emerges: Can effective data-model matching be achieved without direct contact ? Previous methodologies can partially alleviate but not eliminate the necessity of direct contact between data and models, making security and privacy challenges persist throughout the matching process. In this article, our research findings indicate that, despite the essential differences between data and models, both can be represented using topological spaces. Therefore, we establish a unified metric of data complexity and model expressivity from a topological perspective. The unified metric satisfies three conditions toward contactless data-model matching. Then, we develop a contactless matching paradigm, circumventing the necessity for direct contact between data and models and addressing privacy and security concerns. Specifically, we use topological data analysis to generate the data complexity topological descriptors (DCTDs) and use topological simplification to generate the model expressivity topological descriptors (METDs). We compute the matching degree and return the matching result. Through theoretical proof and experimental analysis, we validate the feasibility of the proposed contactless data-model matching paradigm in real-world scenarios.