Research On Representation Traffic Classification Based On Auto-ML

We live in a world with a rapidly progressing technique of Network,it brings us something convenience,it also brings hackers some new way to get through.Firewall,as the first line of defense,supports our cybersecurity and help us resistance against cyber attack.With the development of Artificial Intelligence,the Intelligent state packet detective firewall shows more advantages.The intelligent state packet detection firewall benefits from the development of traffic classify and identify technology.Traditional classify technique needs experts' feature engineering: design features by reading papers,or white books.It's tedious work and usually malfunction.After the emergence of traffic classification technology based on representation learning,firewalls learned to extract features from traffic,which free the lots of researchers from busy work.Unfortunately,Artificial Intelligence has many parameters need to be set,it's a new problem for us engineer.This article researched the traffic classification technology based on representation learning and proposes that using Auto-ML to deal with this problem.We designed a new reward function for this special mission.Besides,we expanded the USTC-TF2016 dataset by added 15 kinds of new malware traffic PCAP data.Training with the expanded dataset,we confirmed that the CNN model created by the Auto-ML system has better performance than Le Net-5,and can be applied in various environments.In the end,this paper designs and implements the state packet detection firewall system based on the representation recognition and the generated CNN architecture and simulates the real attack environment to carry out comparative experiments.The experimental results show that the firewall can realize the specific defense function and can recognize and intercept the attack traffic.