Viewing Federated Learning through the Lens of Small Dataism

Zhang Qian

Abstract

With the increasing awareness of privacy protection, large-scale data collection and sharing has become increasingly difficult, leading to a trend of data “islandization”. To address the problem of joint modeling under data privacy protection, federated learning technology trains a joint model by interacting only with model updates instead of the source data, opening up a new paradigm for data joint modeling. It has made in-depth explorations in aspects such as data heterogeneity, communication efficiency, and model privacy security. However, when viewed from a higher-level perspective of privacy-protected joint modeling, the “immaturity” of federated learning technology will be exposed. This article sorts out the needs and difficulties of joint modeling in real scenarios, analyzes the limitations of federated learning, and introduces small dataism as a guiding ideology to solve the problem of privacy-protected joint modeling. On this basis, it discusses future research directions to help researchers break out of the original framework of federated learning and explore a wider range of technical possibilities.

Keywords

Federated learning; Small dataism; Data privacy protection

References

  1. Zhou Chuanxin, Sun Yi, Wang Degang, et al. A Survey on Federated Learning[J]. Journal of Network and Information Security, 2021, 7(5): 77-92
  2. Zhou Chuanxin. Research on Key Technologies of Data Security Sharing for Federated Learning[D]. Zhengzhou: Information Engineering University of Strategic Support Force, 2023.
  3. Chen Lei, Feng Xiaoqing. A Federated Learning Framework Based on Consortium Chain for Big Data Credit[J]. Credit, 2023, 41(7): 36-42
  4. Xu Yingjin. Fifteen Lectures on the Philosophy of Artificial Intelligence[M]. Beijing: Peking University Press, 2021