An adaptive federated learning system for information sharing in supply chains

Information sharing in supply chains can be challenged by privacy concerns. Equating data and information, the existing literature primarily focuses on the incentivisation behind information sharing between firms. The field of AI may bring a new way of looking at this problem by asking the following question: what if we do not share raw data but share learned information from it instead? This raises the next question, with whom and when should supply chain members share information, which we address in this paper. We develop a novel adaptive federated learning approach for the generation and usage of collective knowledge without direct data exchange and test the approach with a use case for collectively predicting supply risk. We propose a privacy-preserving network formation and clustering algorithm, which enables supply chain members to decide when to enter a collective information-sharing network, and how they should form information-sharing teams. Using data from an e-commerce platform, we illustrate how our approach outperforms the suppliers' own prediction models. We further show that clustering suppliers in teams achieves the best performance and converges faster compared to two benchmarks. The heterogeneity of information contribution by firms and those who benefit from collective information also raises important research questions on the role of cooperation in supply chains.