The new technologies, such as the Internet of Things (IoT), smart grids, and Machine-to-Machine (M2M) communication, herald a new era of computing whereby every conceivable object is equipped with, or connected to, a smart device, allowing a series of data and electronic information to be transmitted through the Internet [1,2]. This information is processed intelligently to create new services for users.
The development of the Internet of things is the final process of mass integration of computer technologies, communication technologies and various sectors of the industrial industry. In addition to threats from traditional communication networks (threats of information distortion, repetition, eavesdropping, disclosure), IOT applications face additional security problems at the application level: the use of cloud computing, intellectual property rights, data processing, protection of confidential information, etc.
Due to the lack of professional testing and evaluation system for cloud computing information security products, the basic security of cloud computing information security products cannot be guaranteed. The establishment of a test and evaluation system of cloud computing information security products is provided, and the system is used in the actual product testing, to further promote the development of cloud computing and information security.
The digital revolution has dramatically changed governance, communication, commerce, and industry worldwide. Digital information advent has given the rise of the ability of being accessed without the need for physical presence
Over the last few decades, information systems have become an increasingly integral aspect of organizations. However, when these information systems are compromised, it can be extremely damaging to business.
In adversarial attacks to machine-learning classifiers, small perturbationsare added to input that is correctly classified. The perturbations yieldadversarial examples, which are virtually indistinguishable from theunperturbed input, and yet are misclassified. In standard neural networks usedfor deep learning, attackers can craft adversarial examples from most input tocause a misclassification of their choice. We introduce a new type of network units, called RBFI units, whose non-linearstructure makes them inherently resistant to adversarial attacks. Onpermutation-invariant MNIST, in absence of adversarial attacks, networks usingRBFI units match the performance of networks using sigmoid units, and areslightly below the accuracy of networks with ReLU units. When subjected toadversarial attacks, networks with RBFI units retain accuracies above 90% forattacks that degrade the accuracy of networks with ReLU or sigmoid units tobelow 2%. RBFI networks trained with regular input are superior in theirresistance to adversarial attacks even to ReLU and sigmoid networks trainedwith the help of adversarial examples. The non-linear structure of RBFI units makes them difficult to train usingstandard gradient descent. We show that networks of RBFI units can beefficiently trained to high accuracies using pseudogradients, computed usingfunctions especially crafted to facilitate learning instead of their truederivatives. We show that the use of pseudogradients makes training deep RBFInetworks practical, and we compare several structural alternatives of RBFInetworks for their accuracy.