%0 Journal Article
%T 基于CNN和RNN的自由文本击键模式持续身份认证<br>Free-text keystroke continuous authentication using CNN and RNN
%A 芦效峰
%A 张胜飞
%A 伊胜伟
%J 清华大学学报（自然科学版）
%D 2018
%R 10.16511/j.cnki.qhdxxb.2018.26.048
%X 个人击键节奏模式具有很难被模仿的特点并可以用于身份认证。根据个人自由文本输入时的击键数据可以学习到个人独有的击键模式。基于对用户自由文本击键输入的检测，能够在不影响用户输入的情况下完成对用户身份的持续认证。该文提出将整体击键数据划分成固定长度的击键序列，并且根据击键的时间特征将击键序列中的击键时间数据转化成击键向量。使用卷积神经网络（convolutional neural networks，CNN）加循环神经网络（recurrent neural networks，RNN）的模型进行个人击键向量序列进行学习，用于身份认证。结果表明：模型使用公开数据集进行实验获得最优拒真率（false rejection rate，FRR）为1.95%，容假率（false acceptance rate，FAR）为4.12%，相等错误率（equal error rate，EER）为3.04%。<br>Abstract：Personal keystroke input patterns are difficult to imitate and can be used for identity authentication. The personal keystroke input data for a free-text can be used to learn the unique keystroke mode of a person. Detection based on a user's free-text keystrokes can be used for continuous identity authentication without affecting the user input. This paper presents a model that divides the keystroke data into fixed-length keystroke sequences and converts the keystroke time data in the keystroke sequence into a keystroke vector according to the time characteristics of the keystrokes. A convolutional neural network and a recurrent neural network are then used to learn the sequences of the personal keystroke vectors for identity authentication. The model was tested on an open data set with an optimal false rejection rate (FRR) of 1.95%, a false acceptance rate (FAR) of 4.12%, and an equal error rate (EER) of 3.04%.
%K 身份认证
%K 击键动力学
%K 自由文本
%K 卷积神经网络(convolutional neural networks
%K CNN)
%K 循环神经网络(recurrent neural networks
%K RNN)
%K &lt
%K br&gt
%K authentication
%K keystroke dynamics
%K free-text
%K convolutional neural networks
%K recurrent neural networks
%U http://jst.tsinghuajournals.com/CN/Y2018/V58/I12/1072