Federated Learning for Multi-Domain Transportation Networks

Federated Learning (FL) can potentially break the framework of traditional cross-domain transportation system modeling and enable realistic and timely cross-domain transportation cooperation. The fusion modeling of transportation system in different domains has been hotly discussed in the field of transportation in recent years. However, due to the data conflict problem caused by heterogeneous data and privacy constraints, traditional cross-domain cooperation is delayed or even difficult. The proposed FL framework is suitable for cross-domain modeling problems with non-IID heterogeneous data stored in devices. The novelty of establishing FL in the transportation field is that it connects transportation systems across different domains and builds cross-domain collaborative models without data sharing. FL reshapes the data exchange framework between domains, realizes data privacy protection, and breaks the barrier of direct multi-domain data exchange to a certain extent. The combination of FL and cross-domain transportation modeling enriches FL modeling applications. The model infers and assesses the wide-area transport environment, including both local and susidiary inferred data.