Design method of convolutional input type nested recurrent neural network

Abstract

The invention discloses a design method of a convolutional input type nested recurrent neural network. The design method comprises the following steps of performing data combination and convolution operation processing on current moment input data and last moment output data; equally splitting a convolution result to serve as each gating unit in the original long-term and short-term memory networkunit; carrying out convolution operation in the inner-layer nested unit to serve as input, carrying out gating calculation operation the same as that of the long-term and short-term memory network unit, and obtaining output of the inner-layer nested unit; and taking the output of the inner-layer nesting unit as a memory unit value of the outer-layer unit, and obtaining a final output value of the whole unit through an output gate. According to the method, the combination of a nested recurrent neural network and convolution input is provided, so that the performance of the model for fitting long-time associated data is improved, the local association between feature relationships is extracted, and a certain parameter quantity is reduced; and compared with a common recurrent neural network, the method has higher accuracy and fewer parameters.

Description

technical field [0001] The invention belongs to the technical field of data processing, and relates to a design method of a recursive neural network unit, in particular to a design method of a convolution input nested recursive neural network. Background technique [0002] With the development of network technology and hardware in the Internet of Things era, the number of users and connected devices has exploded. In 2017, the number of IoT devices exceeded the global population of 7.5 billion for the first time, and by 2020, this number is expected to grow to more than 30 billion. However, the application of IoT devices is still in its infancy, and it still faces many security risks due to the large number of devices and the simple structure of their own. The most important of these is the security issue of malicious intrusion from the Internet. [0003] Recurrent neural networks are widely used in the prediction of time series; for intrusion attacks from the network, the ...

AI Technical Summary

Problems solved by technology

[0010] Purpose of the invention: In order to overcome the problems of poor sensitivity to long-term sequences and low test accuracy when existing recurrent neural networks model network intrusion data in the prior art, a nested recursive neural network with convolutional input is provided. The network design method, on the basis of solving these problems, reduces the number of parameters used to a certain extent, and takes into account the local correlation between features, improving the test accuracy

Images