ICMFed: An Incremental and Cost-Efficient Mechanism of Federated Meta-Learning for Driver Distraction Detection

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