【摘要】 设计一种PE格式恶意软件混淆对抗样本生成模型。利用深度强化学习算法,实现对恶意软件的自动混淆。通过加入历史帧和LSTM神经网络结构的方法使深度强化学习模型具有记忆性。对比实验表明,该恶意软件变种在基于机器学习的检测模型上的逃逸率高于现有研究,在由918个PE格式恶意软件组成的测试集上达到39.54%的逃逸率。更多还原

【Abstract】 This paper designs a PE malware obfuscation adversarial sample generation model. It used deep reinforcement learning algorithms to realize automatic obfuscation of malware, and the deep reinforcement learning model was provided with memory through historical frames and LSTM neural network structure. Comparative experiments show that the escape rate of the generated malware variants on the machine learning-based detection model is higher than that of existing research, reaching 39.54% on the test set composed of 918 PE malware.更多还原