2. With the growing adoption of encrypted traffic, Encrypted Network Traffic
Classification of different sorts has been very pivotal to network management,
visibility and security. With scaling networks and network usage the
continuous monitoring of networks has become very important.
We observe the flow statistical properties from the packet header. We
specifically propose using a residual network to treat statistical network data as
images and perform classification.
3. Solves the vanishing gradient problem in deep
neural networks.
Deeper neural networks are better at classifying
highly complex data.
Gives us a general classification framework with
networking traffic modelling instead of using
specific models for specific use cases.
Ability to model complex non-linear relations
and achieve more expressiveness and
generalization.
Better scalability