Files and code for 'Genetically Optimized Massively Parallel Binary Neural Networks for Intrusion Detection Systems' T.Murovič, A.Trost
Original datasets are taken from:
- https://www.unsw.adfa.edu.au/unsw-canberra-cyber/cybersecurity/ADFA-NB15-Datasets/ [UNSW-NB15 dataset]
- https://www.unb.ca/cic/datasets/nsl.html [NSLKDD dataset]
They are also tagged at 
The cybersecurity_dataset_x.m scripts take the original datasets and binarize them and format them as explained in the paper. The outputs are also available at
BNNs are then trained on the new datasets using the scripts from 
Net_Params/ --> Includes the trained BNN parameters per layer for the UNSW-NB15 and NSLKDD dataset, both in the original partition (FULL) and the new, randomly permutted partition (RAND). (_w are binary weights, _t are integer comparison thresholds).
Scripts build_layers.m and hamming_genetic.m are then used to automatically build the optimized BNN combinational HDL (Verilog) modules.
Verilog/ --> Includes the Verilog modules of BNN layers for the UNSW-NB15 (dataset_0) and NSLKDD (dataset_1) dataset, both in the original partition (partition_1) and the new, randomly permutted partition (partition_0). Can be used as is in the desired HDL project.