In Patterns (New York, N.Y.)
In the big data era, vast volumes of data are generated daily as the foundation of data-driven scientific discovery. Thanks to the recent open data movement, much of these data are being made available to the public, significantly advancing scientific research and accelerating socio-technical development. However, not all data are suitable for opening or sharing because of concerns over privacy, ownership, trust, and incentive. Therefore, data sharing remains a challenge for specific data types and holders, making a bottleneck for further unleashing the potential of these "closed data." To address this challenge, in this perspective, we conceptualize the current practices and technologies in data collaboration in a data-sharing-free manner and propose a concept of the model-sharing strategy for using closed data without sharing them. Supported by emerging advances in artificial intelligence, this strategy will unleash the large potential in closed data. Moreover, we show the advantages of the model-sharing strategy and explain how it will lead to a new paradigm of big data governance and collaboration.
Li Zexi, Mao Feng, Wu Chao
2022-Nov-11
artificial intelligence, big data, data sharing, federated learning, model-sharing strategy, open science
