Federated learning in food research
Zuzanna Fendor, Bas H. M. van der Velden, Xinxin Wang, Andrea Jr. Carnoli, Osman Mutlu, Ali Hürriyetoğlu
Abstract
The use of machine learning in food research is sometimes limited due to data sharing obstacles such as data ownership and privacy requirements. Federated learning is a technique to potentially alleviate these obstacles because it allows to train machine learning models locally, keeping the data private and sharing only the learned parameters. In this review we investigate the use of federated learning in food research. First, we outline a framework that describes the variants of federated learning implementations. Then, we provide an overview of applications of federated learning in food research. Next, we discuss the performance of the models trained with federated learning, and reasons for the use of federated learning. Finally, we categorize the encountered federated learning applications within the federated learning framework. In the discussion, we highlight the knowledge gaps and discuss the potential novel applications. In this review we examined a total of 86 papers published between 2019 and 2024. The current applications encompass crop disease monitoring, yield prediction, quality assessment, and pesticide residue risk analysis. We observed the general trend of centralized horizontal federated learning, and identified the absence of vertical federated learning, federated transfer learning, and decentralized architectures as research gaps.