Hardness of evasion of multiple classifier system with non-linear classifiers
Fei Zhang, Wei Jie Huang, Patrick P. K. Chan · 2014
Many studies have shown that Multiple Classifier Systems (MCSs) are more robust than single classifiers to evasion attacks for linear classifiers. However, to the best of our knowledge, the robustness of MCSs for non-linear classifiers has not been inves-tigated. This paper attempts to discuss two issues experimentally including a MCS is still more robust than a single classifier for non-linear classifiers, and also a non-linear classifier is more robust than a linear classifier. Besides the accuracy, we adopt the hardness of evasion as the evaluation criterion to measure the robustness of a classifier. The results show that MCSs and non-linear classifiers are more robust to the evasion attack generally.