A Machine Learning Accelerator for DDoS Attack Detection and Classification on FPGA

This paper presents the hardware accelerators to detect and classify DDoS attacks. Two types of neural network models based on different observation methods and features are demonstrated to be capable of processing packet streams at l00Gbps wire speed with good performance. The CIC2019 DDoS dataset is used and processed utilizing data augmentation techniques with enough variation for training the machine learning models. Finally, we discussed the cost of implementing these two models on the Xilinx Alveo U200 Data acceleration card.