Secure Deep Learning in Defense in Deep-Learning-as-a-Service Computing Systems in Digital Twins
Zhihan Lv, Dongliang Chen, Bin Cao, Houbing Song, Haibin Lv
Abstract
While Digital Twins (DTs) bring convenience to city managers, they also generate new challenges to city network security. Currently, cyberspace security becomes increasingly complicated. Intrusion detection and Deep Learning (DL) are combined with shunning security threats in service computing systems and improving network defense capabilities. DTs can be applied to network security. People's understanding of cyberspace security can be improved using DTs to digitally define, model, and display the network environment and security status. The intrusion detection data are optimized based on DL technology, and a network intrusion detection algorithm integrated with Deep Neural Network (DNN) model is proposed. In the cloud service system, a trust model based on Keyed-Hashing-based Self-Synchronization (KHSS) is introduced. This model predicts the security state and detects attacks according to existing malicious attacks, ensuring the network security defense system's regular operation. Finally, simulation experiments verify the Deep Belief Networks (DBN) model's feasibility and the cloud trust model. The DBN algorithm proposed improves the correct detection rate of unknown samples by 4.05% compared with the Support Vector Machine (SVM) algorithm. From the 20,100 pieces of data in the test dataset, the number of correct attacks detected by the DBN algorithm exceeds those by the SVM algorithm by 818. DBN algorithm requires a short detection time while ensuring optimal detection accuracy. The KHSS+DBN model predicts cloud security states, and the results are the same as the actual states, with an error of only 1%∼2%.