Using LSTM Cells for SIP Dialogs Mapping and Security Analysis
Jan Rozhon, Erik Gresak, Jakub Jalowiczor · 2018
Past decade has witnessed a rapid growth of a number of use cases employing the neural networks for many diverse purposes starting with user identification through anomaly detection and ending with speech synthesis. This growth is bolstered by the increasingly accessible massive computational power provided by both modern processors and graphic processing units (GPUs), community driven rapid development of machine learning frameworks such as Tensorflow or Theano, and the ability of neural networks to successfully find the relation that is difficult if not impossible to obtain analytically. In the field of modern communications, the anomaly detection algorithms have been a focus of many scientific and industrial studies since the late detection of attack/misconfiguration can result in massive economic losses in telecommunications industry. This paper follows the pattern of evolution of artificial intelligence algorithms employed in the telecommunications field and moves the field further by considering the signaling exchange as simple language protocol and creating a model that can learn a dialect of individual nodes in the network and detect if this dialect changes and thus alerting the network administrator of a possible break in. The abstraction of the signaling protocol used allows for usage of recurrent neural networks or more specifically Long-Short Term Memory (LSTM) Cells that have never been used for this purpose as far as the authors can tell.