The Simulation Study on the Effect of Transmission Error in Split Computing Approach

The split computation service has gained wide attention to reduce the artificial intelligence-based service completion time. In a split computation service, the mobile device and edge server need to exchange the intermediate data to cooperatively process the AI-based service (i.e., process the deep model in AI-based service). In this case, the performance of split computation (e.g., classification accuracy) is degraded due to the transmission error. However, most of the works did not consider the transmission error effect of the split computation service. In this paper, we conduct various simulations to analyze the effect of transmission error on the split computation service. From the simulation results, we find some general trends that the later deep layer and convolution layer have better error-tolerant capability.