基于元学习的入侵检测研究

Research of Intrusion Detection Based on Meta-Learning

入侵检测可以防御网络的攻击,但现有方法存在需要大量训练样本、泛化能力低、效果欠佳等不足,因此文章提出一种基于带记忆池的模型无关的元学习方法(Model-Free Meta-Learning with Memory,MFMLM)。MFMLM算法采用带记忆池的模型无关的元学习方法处理网络入侵检测问题,利用多轮采样获得多个小样本数据集,通过在多个数据集中进行训练,得到最优参数。方法利用元学习改善模型的判别和泛化能力,有效地缩短了训练时间,提高了检测准确率。实验结果表明,所提方法可以较好地解决入侵检测问题,具有良好的性能。

Although many current intrusion detection algorithms guard against network attacks,they require tons of training samples and suffers from low generalization capability and unsatisfactory performance.To solve this problem,we propose a meta-learning-based,model-free intrusion detection algorithm,referred as Model-Free Meta-Learning with Memory(MFMLM),which takes advantage of a memory pool.The MFMLM algorithm produces small sample data sets by sampling tasks several rounds,and trains the model with the data sets in search of optimal parameters.With the introduction of meta-learning,the discrimination and generalization capacity of the model is improved,so are the training speed and detection accuracy.The experiment results indicated that MFMLM algorithm could efficiently handle intrusion detection with satisfying performance.