Splitfed learning for smart agriculture

Split learning emerges as a promising approach for edge computing in resource-constrained contexts, with improved data privacy. However, SL’s client-side model synchronisation, which occurs sequentially, causes additional delay in model training/testing. This latency presents issues, particularly in industries such as health and finance, where rapid model updates are critical for anomaly and fraud detection in real time. Splitfed learning is proposed to overcome these difficulties while also leveraging the benefits of federated learning and split learning. Splitfed learning refers to a novel federated learning strategy that distributes model training across several edge devices while maintaining data privacy and reducing communication overhead. Splitfed learning aids smart agriculture by allowing for collaborative model training, privacy preservation and scalability. Decentralised data ownership, split model design and secure communication make it easier to create strong and accurate global models. However, difficulties such as communication overhead, model synchronisation and security concerns must be solved before deployment can be completed successfully. Future directions are optimising communication protocols, improving model synchronisation mechanisms and combining splitfed learning with edge computing to promote innovation in sustainable farming methods and food security.