The evolution of intelligent transportation systems is enabling a shift toward distributed, data-driven smart mobility, where vehicles and infrastructure collaboratively learn from continuously generated data. Distributed AI plays a key role in this transformation by enabling learning directly within vehicular environments while addressing challenges such as privacy, scalability, bandwidth efficiency, and latency.

This keynote presents recent advances in communication-efficient and privacy-preserving distributed learning for vehicular networks. It focuses on decentralized learning paradigms, including Federated Learning and gossip-based model exchange, where vehicles and infrastructure collaboratively train models without sharing raw data. Emphasis is placed on efficient communication strategies such as layer-wise update selection and partial model sharing, which reduce communication overhead while maintaining model performance.

The talk highlights how these techniques enable scalable collaboration in dynamic mobility environments and discusses representative applications such as driver behavior profiling, anomaly detection, and safety-critical decision-making. Overall, the keynote provides a unified view of distributed AI for smart mobility, focusing on efficient collaboration, privacy preservation, and practical deployment at the edge.