A Novel Approach of Unknown Network Attack Detection Based on Zero-Shot Learning
Hui Wang, Yifeng Wang, Yuanbo Guo · 2021 IEEE International Conference on Data Science and Computer Application (ICDSCA) · 2021
Along with the computer network and software systems tend to be complex, using new technology, the method such as holes or social engineering skills on the implementation of the unknown network attack methods emerge in endlessly. Due to lack of corresponding labels in such attacks when detecting actually, traditional learning methods which need large amounts of labelled data will often appear low accuracy, besides, existing approaches of zero-shot learning suffer from domain shift problems. In our paper, we put forward a novel approach of unknown network attack detection on the basis of zero-shot learning. Under the multi-metric space, we align semantic embedding and visual embedding via designing a zero-shot learning model based on variational autoencoder, and transfer known class learning to the unknown class through the semantic feature vector. This approach not only alleviates the projection domain shift problem, but also improving the detection accuracy of unknown network attacks. Experiments show that our model has good practicability and feasibility on the NSL - KDD dataset, comparing with the other classifier models.