project has been narrowed down once again hopefully to a final idea :)

the idea is to use recurrent neural networks (possible multiple) in conjunction with decision tree methods in order to create the red teaming software

the software would be able to scan an entire network (maybe cluster nodes into sections - kmeans) in order to select which system to attack. This is most likely to be done with nessus if i can use the universities license for my project. If not then openVAS is the next choice. the software would then use a decision tree (with a RNN inside) to plot an attack against the targeted system. it would execute this attack using metasploits CLI, possibly getting the exploit from the nessus output

Theano is the framework on choice and the bulk of the project will most likely be written in python

Ive yet to further expand on this and create a title which is worrying since the proposal deadline is just over a week away.

in terms of practical work, Ive managed to get Theano running on my overclocked 680 on Windows 10. This was much more of a challenge than it should have been.

im not sure whether or not the system will be using windows or linux simply due to the compatibility of the software.

the output of the scanner would need to be converted into some format that the RNN would understand and im not sure how i will be able to train the model on exploiting machines. For a simple POC the project could run against 10 vulnerable VM’s.