A0501
Title: A comparative study on federated learning in supply chain forecasting: Pareto optimality on accuracy and efficiency
Authors: Hang Qi - BNU-HKBU United International College (China)
Jieping Luo - BNU-HKBU United International College (China)
Jingjin Wu - BNU-HKBU United International College (China) [presenting]
Abstract: Federated learning (FL) is an emerging learning mechanism that can achieve accuracy, efficiency, and privacy concurrently, which could be particularly useful in scenarios where large volumes of sensitive data are involved, such as supply chain management and forecasting. A dataset composed of sales records of multiple products across five different regions globally is considered, and a comparative study of centralized learning, distributed learning, and FL focuses on the accuracy of predicting future demand for certain products and the required volume of data for transmission by the multi-layer perception (MLP) model. The results show that MLP with FL has the fastest convergence rate; that is, it can achieve the highest accuracy in predicting the future demands of certain products compared to the other two learning approaches, given the same number of training rounds. In addition, the performances of FL approaches are compared with different combinations of market data sets across regions, and the Pareto optimality is examined in terms of accuracy and transmission efficiency under different scenarios. The potential transformative impact of FL is demonstrated on global supply chain management in various aspects.
