Few-shot malicious traffic classification based on Siamese Neural Network
Kailin Wu, Pan Wang, Zixuan Wang · 2021
The process of learning good features for machine learning applications can be computationally expensive, and can prove difficult when there is little data available. A typical example is few-shot learning. In this case, we must make predictions correctly and only give a few examples of each class. In this article, we propose a method based on a one-dimensional convolutional siamese neural network, which uses a unique structure to naturally rank the similarities between inputs. Once the network is adjusted, we can use powerful discriminative features to extend the network's predictive capabilities to new data, and to new categories from unknown distributions. Using the convolutional architecture, we can achieve powerful results that exceed other deep learning models, and have close to the most advanced performance on few-shot classification tasks. The results show that this method can obtain higher recognition accuracy than the traditional methods of smote and GAN generated data training.