No abstract

Looking ahead to the future-stage autonomous transportation system (ATS), personal mobility service (PMS) aims to provide the recommended travel options based on both microscopic individual travel demand and the macroscopic supply system objectives. Such a goal relies on massive heterogeneous data to interpret and predict user travel intentions, facing the challenges caused by prevalent centralized approaches, such as an unbalanced utilization rate between cloud and edge, and data privacy. To fill the gap, we propose a federated logit model (FMXL), for estimating user preferences, which integrates a discrete choice model—the mixed logit model (MXL), with a novel decentralized learning paradigm—federated learning (FL). FMXL supports PMS by (1) respectively performing local and global estimation at the client and server to optimize the load, (2) collaboratively approximating the posterior of the standard mixed logit model through a continuous interaction mechanism, and (3) flexibly configuring two specific global estimation methods (sampling and aggregation) to accommodate different estimation scenarios. Moreover, the predicted rates of FMXL are about 10% higher compared to a flat logit model in both static and dynamic estimation. Meanwhile, the estimation time has been reduced by about 40% compared to a centralized MXL model. Our model can not only protect user privacy and improve the utilization of edge resources but also significantly improve the accuracy and timeliness of recommendations, thus enhancing the performance of PMS in ATS.

Driver distraction detection (3D) is essential in improving the efficiency and safety of transportation systems. Considering the requirements for user privacy and the phenomenon of data growth in real-world scenarios, existing methods are insufficient to address four emerging challenges, i.e., data accumulation, communication optimization, data heterogeneity, and device heterogeneity. This paper presents an incremental and cost-efficient mechanism based on federated meta-learning, called ICMFed, to support the tasks of 3D by addressing the four challenges. In particular, it designs a temporal factor associated with local training batches to stabilize the local model training, introduces gradient filters of each model layer to optimize the client–server interaction, implements a normalized weight vector to enhance the global model aggregation process, and supports rapid personalization for each user by adapting the learned global meta-model. According to the evaluation made based on the standard dataset, ICMFed can outperform three baselines in training two common models (i.e., DenseNet and EfficientNet) with average accuracy improved by about 141.42%, training time saved by about 54.80%, communication cost reduced by about 54.94%, and service quality improved by about 96.86%.

No abstract

Driven by technologies and demands, the modern transportation system has developed from intelligent transportation systems (ITS) to autonomous transportation systems (ATS) to resolve intertwined demands and supplies with few human interventions. In ATS, personal mobility service (PMS) is the service that can sense real-time traffic conditions comprehensively, learn travelers’ preferences accurately, recommend multimodal travel options appropriately, and provide service responses timely to elevate the level of personalization and intelligence in smart mobility services. Since current PMS widely employs centralized approaches (CPMS) to process massive sensitive data from individuals and support diverse edge devices, resulting in high pressure in privacy protection and performance balancing, this paper presents a federated PMS (FPMS) and its design architecture in logical and physical views by adopting federated learning to provide multimodal, dynamic, and personalized travel options with system-saving safety and efficiency guaranteed. Moreover, through an extensive evaluation, the performances of CPMS and FPMS are compared to reveal the merits of FPMS in reducing costs and latency.