This paper presents a new hard-wired combinational Binary Neural Network (BNN) architecture and its design methodology, where the networks are constructed and trained with the aim of tackling the problem of classifying IP packets efficiently for Intrusion Detection Systems (IDSs). Shallow, single-hidden-layer BNNs are trained on benchmark NSL-KDD and UNSW-NB15 datasets and achieve accuracy rates (77.77% to 98.96%) comparable to those of similar compact networks used for detecting intrusions. These networks are then implemented in Field-Programmable Gate Arrays (FPGAs) using this novel combinational ripple architecture, which is optimized using a genetic algorithm and uses neuron-to-neuron similarities to achieve state-of-the-art performance in terms of resource usage (8606 to 17990 lookup tables) and classification latency (16–$19\mathrm{ns}$).

本文提出了一种新的硬连线组合二元神经网络 (BNN) 架构及其设计方法，其中构建和训练网络的目的是解决入侵检测系统 (IDS) 的 IP 数据包有效分类问题。浅层单隐藏层 BNN 在基准 NSL-KDD 和 UNSW-NB15 数据集上进行训练，其准确率（77.77% 至 98.96%）可与用于检测入侵的类似紧凑网络相媲美。然后使用这种新颖的组合纹波架构在现场可编程门阵列 (FPGA) 中实现这些网络，该架构使用遗传算法进行优化，并使用神经元到神经元的相似性在资源方面实现最先进的性能使用率（8606 到 17990 查找表）和分类延迟（16–$19\mathrm{纳秒}$）。