原文献名称 Predicting Domain Generation Algorithms with Long Short-Term Memory Networks
原文作者 Jonathan Woodbridge, Hyrum S. Anderson, Anjum Ahuja, and Daniel Grant
The LSTM cell’s design with multiplicative gates allows a network to store and access state over long sequences, thereby
mitigating the vanishing gradients problem. For our use with domain names, the state space is intended to capture combi-
nations of letters that are important to discriminating DGA domains from non-DGA domains. This flexible architecture
generalizes manual feature extraction via bigrams, for example, but instead learns dependencies of one or multiple characters,
whether in succession or with arbitrary separation.
LSTM网络的多重门结构使得LSTM网络可以存储并且获取长序列的状态，因而减轻了特征失效这一个问题。在域名检测这一问题
上，我们主要利用LSTM的状态空间来存储字母之间的综合特征，这些状态是将DGA域名区分出来的关键。这样存储的综合特征覆盖
范围之广，包括绝大多数手工创建的特征，例如：依赖于一个字母或者多个字母的连续特征或者间断特征。