Deep Learning Applications for Cyber Security by Unknown

Author:Unknown
Language: eng
Format: epub
ISBN: 9783030130572
Publisher: Springer International Publishing

5.2 1D-CNN

1D-CNNs have been typically used in NLP tasks where information can be represented as a sequential one-dimensional array. Although there are two representative dimensions in the Flow-Image input, the bytes between a packet have a higher degree of correlation than the bytes over multiple packets. A 1D-CNN could be used to exploit this property of network traffic. A diagram of the 1D-CNN architecture is shown in Fig. 4.

5.3 Segmented-CNN

A Segmented-CNN is a novel CNN architecture that aims to capitalise on the distinct properties of the two sections in a TCP/IP packet, the header and the payload. This architecture is inspired by the approach taken by Bromley et al. in their design of a Siamese-CNN [24]. A Siamese-CNN utilises two CNNs to simultaneously analyse two distinct images, the output of which is then compared to determine the similarity between the two. However, instead of measuring the similarity between these two images, this methodology could be adapted to analyse two inputs that relate to the same classification but have their own distinct properties. Figure 5 shows how two CNNs could be used to analyse the distinct sections of a TCP/IP connection.

Fig. 5Structure of a Segmented-CNN showing how the two CNN networks can be individually optimised for the two distinct regions of a packet, the header and payload. Only the first convolutional, pooling, and fully-connected layers of the Segmented-CNN have been shown

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.