Abstract
Deep learning model shows great advantages in various fields. However, researchers pay attention to how to improve the accuracy of the model, while ignoring the security considerations. The problem of controlling the judgment result of deep learning model by attack examples and then affecting the system decision-making is gradually exposed. In order to improve the security of sentence similarity analysis model, we propose a convolution neural network model based on attention mechanism. First of all, the mutual information between sentences is correlated by attention weighting. Then, it is input into improved convolutional neural network. In addition, we add attack examples to the input, which is generated by the firefly algorithm. In the attack example, we replace the words in the sentence to some extent, which results in the adversarial data with great semantic change but slight sentence structure change. To a certain extent, the addition of attack example increases the ability of model to identify adversarial data and improves the robustness of the model. Experimental results show that the accuracy, recall rate and F1 value of the model are due to other baseline models.
Highlights
Deep learning model shows great advantages in various fields, including computer vision, classification system, prediction model and so on [1]–[6]
THE PROPOSED MODEL In this paper, we propose an adversarial convolution neural network model based on attention mechanism, which can deal with the security of adversarial modulation by adding adversarial mechanism to convolution neural network
MODEL TRAINING we introduce the training process of anticonvolution neural network based on multi feature attention mechanism
Summary
Deep learning model shows great advantages in various fields, including computer vision, classification system, prediction model and so on [1]–[6]. In the task of sentence similarity analysis, the intrusion of adversarial sample always imposes some key security obstacles and attacks the model Based on this very situation, we propose a multi feature model to extract feature information from sentences. The interactive adversarial examples generation method can ensure the change rate of samples and the performance of the model, which is of great significance for training high security model in practical application. The third chapter is the proposed model, including multi feature attention model, the generation of adversarial examples based on genetic algorithm and the counter convolution neural network model. We verify the performance of the model through experiments, including the accuracy of sentence similarity analysis and the security of dealing with adversarial examples.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have