Privacy and Security in Vehicular Data Aggregation Using Federated Learning - A Parking Occupancy Case Study

Explore a 16-minute conference talk examining privacy concerns in vehicular federated learning systems for future smart cities. Dive into the potential privacy leakage risks associated with model updates and gradient sharing, where curious servers could potentially infer vehicle routes and private information. Through a simulated scenario, learn about vulnerability patterns in vanilla federated learning that expose both intra-city and commuter traffic to location inference attacks. Discover how adversarial aggregator servers can successfully determine vehicle movement times during low-traffic periods, highlighting critical privacy challenges in intelligent transportation systems. Presented by Levente Alekszejenkó, this Eclipse Foundation talk provides valuable insights into balancing the benefits of federated learning with essential privacy considerations in modern urban mobility.