Semantic Model Component Implementation for Model-driven Semantic Communications

TL;DR

The cross-source-domain and cross-task semantic component model is designed, which uses smaller model parameters to achieve cross-source, cross-task functionality while maintaining performance and improving the model's tolerance to noise.

Abstract

The key feature of model-driven semantic communication is the propagation of the model. The semantic model component (SMC) is designed to drive the intelligent model to transmit in the physical channel, allowing the intelligence to flow through the networks. According to the characteristics of neural networks with common and individual model parameters, this paper designs the cross-source-domain and cross-task semantic component model. Considering that the basic model is deployed on the edge node, the large server node updates the edge node by transmitting only the semantic component model to the edge node so that the edge node can handle different sources and different tasks. In addition, this paper also discusses how channel noise affects the performance of the model and proposes methods of injection noise and regularization to improve the noise resistance of the model. Experiments show that SMCs use smaller model parameters to achieve cross-source, cross-task functionality while maintaining performance and improving the model's tolerance to noise. Finally, a component transfer-based unmanned vehicle tracking prototype was implemented to verify the feasibility of model components in practical applications.

Figures (10)

• Figure 1: Layer-based image semantic communication system consists of basic model and enhancement models. Each enhancement model can be regarded as a semantic model component (SMC) that can control accuracy and semantics.
• Figure 2: (a) Illustration of Incremental SMC framework. For image classification tasks, incremental SMC of the corresponding classification categories are added to the base model to increase the classification categories of the model. (b) Illustration of Cross-source-domain SMC framework. For the semantic segmentation task, the base model is used to perform semantic segmentation of the gta5 dataset, which is a virtual view of cars on the city streets. By adding a cross-source-domain SMC of the cityscapes dataset, the model has the ability to segment the cityscapes dataset, enabling cross-modality from virtual new data sources to real-world data sources. (c) Illustration of Cross-task SMC framework. Add cross-task SMC to the base model to give the original model the ability to perform additional tasks.
• Figure 3: Model architecture of expandable representation learning. Base model is composed of a feature extractor $\Phi = [\Phi_G, \Phi_S]$ and the task execution module $\mathcal{H}$. The generalized feature extractor $\Phi_G$ will be reused and the special feature extractor $\Phi_S$ will be expanded using the new special feature extractor $\Phi_{S_{new}}$. In addition, the task execution model structure and parameters need to be updated for cross-task SMC.
• Figure 4: (a) The structure of classification basic model and Incremental model component. (b) The structure of Classification basic model and cross-task SMC. (c) The structure of Detection basic model and cross-task SMC. (d) The structure of Segmentation basic model and cross-source-domain SMC.
• Figure 5: The influence of the incremental SMC performance with different signal-to-noise ratio under the condition of the second-order loss regularization term.