Deep Semantic Communication for Knowledge Sharing in Internet of Vehicles

Along with the development of intelligent transportation system (ITS), artificial intelligence (AI)-based machine learning technologies have been widely utilized in Internet of Vehicles (IoV). Neural network (NN)-based knowledge sharing among vehicles and road side units (RSUs) presents considerable benefits for enhancing vehicle intelligence. However, it is challenging to ensure the efficiency of knowledge sharing under unstable connectivity among vehicles with different NN model architectures. In this article, we propose a new deep semantic communication framework for knowledge sharing (SCKS), enabling one-to-many NN model transmission and realizing efficient knowledge sharing in an IoV. Based on this framework, a generative distillation algorithm is designed to extract the semantic features of NN model, which can ensure the efficiency of the transmitter for knowledge sharing across different NN models and reduce communication bandwidth demand. In order to facilitate an effective understanding of semantic information by heterogeneous receivers, we design a generative adversarial networks (GAN)-based semantic decoding algorithm. Numerical results on CIFAR10 and ImageNet datasets show that the proposed SCKS outperforms the baseline, especially in the low-signal-to-noise (SNR) region. In particular, the simulation results demonstrate superiority of proposed SCKS scheme in terms of bandwidth requirements and computational efficiency for knowledge sharing cross different NN architectures than the state-of-art scheme, including DeepJSCC and knowledge distillation (KD).