Multi-layer model and training method for information-extreme malware traffic detector
Альона Сергіївна Москаленко, В’ячеслав Васильович Москаленко, Артур Фанісович Шаєхов, Mykola Zaretskyi · 2020
Model-based on multilayer convolutional sparse coding feature extractor and information-extreme decision rules for malware traffic detection is presented in the paper.Growing sparse coding neural gas algorithms for unsupervised pre-training of the feature extractor are used.Random forest regression model as a student in knowledge distillation from sparse coding layers is proposed for speed up inference mode.Information-extreme learning method based on binary encoding with tree ensembles and class separation with radial basis function in binary Hamming space are proposed.Information-extreme classifier is characterized by low computational complexity and high generalization ability for small labeled training sets.Simulation results with an optimized model on test open datasets confirm the suitability of proposed algorithms for practical application.