Generative Pretrained Transformer for Network Traffic

View the PDF file for the paper entitled NetGPT: a indelible transformer for the movement of the network, by Xuying Meng and 3 other authors

PDF HTML (experimental) view

a summary:All data on the Internet is transferred by network traffic, and therefore network movement modeling can help improve the quality of network services and protect data privacy. The pre -models of the network traffic can be widely used the initial data to learn the basic properties of the network traffic, and to create discriminatory results for the input traffic without considering the specified estuary tasks. Prepared effective models can improve the efficiency of training and the effectiveness of clinic tasks significantly, such as classification of applications, detection of attack and traffic generation. Despite the great success in training in the treatment of natural language, there is no work in the network. Looking at the various requirements, network traffic properties and network tasks, it is not distinguished to build a pre -movement model for the network traffic and we face various challenges, especially heterogeneous heads and the load in the movement of multiple models and different consequences of various contexts of networks below the river.

To deal with these challenges, in this paper, we are making a first attempt to provide a model NetGPT model for each of the tasks of understanding traffic and generation tasks. We suggest multi -mode network traffic modeling to build uniform text inputs and support each of the tasks of understanding traffic and its generation tasks. We also improve the effect of adaptation to the pre -model of various tasks by mixing the headfields, retaining packages in flows, and merging various task stickers with claims. Through various traffic data collections of encrypted programs, DNS, private industrial protocols, and mining in the encrypted currency, expensive experiences show our NetGPT event in a set of traffic understanding tasks and traffic generation tasks, and outperform the latest dishes in the latest model.