Nowadays, deep learning has been increasingly applied in real-world scenarios involving the collection and analysis of sensitive data, which often causes privacy leakage. Differential privacy is widely recognized in the majority of traditional scenarios for its rigorous mathematical guarantee. However, it is uncertain to work effectively in the deep learning model. In this paper, we introduce the privacy attacks facing the deep learning model and present them from three aspects: membership inference, training data extraction, and model extracting. Then we recall some basic theory about differential privacy and its extended concepts in deep learning scenarios. Second, in order to analyze the existing works that combine differential privacy and deep learning, we classify them by the layers differential privacy mechanism deployed, such as input layer, hidden layer, and output layer, and discuss their advantages and disadvantages. Finally, we point out several key issues to be solved and provide a broader outlook of this research direction.

Automatic modulation classification (AMC) is one of the most essential algorithms to identify the modulation types for the non-cooperative communication systems. Recently, it has been demonstrated that deep learning (DL)-based AMC method effectively works in the single-input single-output (SISO) systems, but DL-based AMC method is scarcely explored in the multiple-input multiple-output (MIMO) systems. In this correspondence, we propose a convolutional neural network (CNN)-based cooperative AMC (Co-AMC) method for the MIMO systems, where the receiver, equipped with multiple antennas, cooperatively recognizes the modulation types. Specifically, each received antenna gives their recognition sub-results via the CNN, respectively. Then, the decision maker identifies the modulation types, based on these sub-results and cooperative decision rules, such as direct voting (DV), weighty voting (WV), direct averaging (DA) and weighty averaging (WA). The simulation results demonstrate that the Co-AMC method, based on the CNN and WA, has the highest correct classification probability in the four cooperative decision rules. In addition, the CNN-based Co-AMC method also performs better than the high order cumulants (HOC)-based traditional AMC methods, which shows the effective feature extraction and powerful classification capabilities of the CNN.

In the cloud-assisted Industrial Internet of Things (IIoT), ciphertext-policy attribute-based encryption (CP-ABE) could help the data owner (DO) share his sensitive data via the cloud under self-defined access structures. Among general CP-ABE schemes, the decryption overhead, the key generation cost, and the ciphertext length increase with the number of involved attributes. Additionally, only regular attributes are taking into consideration rather than weighted attributes. In this article, we proposed a secure, efficient, and weighted access control scheme (SEWAC) for cloud-assisted IIoT applications. SEWAC enables the DO to formulate any fine-grained access structure over weighted attributes without making it more complicated. Furthermore, such weighted attributes would not add the length of ciphertext. SEWAC also supports online/offline key generation to alleviate the computational cost of the authority from answering mass key requests in the online phase, while most computational tasks are executed in the offline phase. The heavy decryption overhead is offloaded to the cloud. To ensure the cloud to honestly execute the process of outsourced decryption, we design an efficient batch verification method, which allows the user to spend only three bilinear pairing operations in checking the correctness of batch results. We also give the formal security proof of the proposed scheme. Comprehensive comparisons and implementation results indicate that SEWAC can better achieve weighted access control, compressed ciphertext length, efficient key generation, and the assurance of the outsourced decryption result.

Many fields, such as remote sensing, medical imaging, and biological detection, pose a technical challenge for achieving super-resolution imaging. Convolutional neural networks (CNNs) are considered one of potential solutions to realize the super-resolution. In this paper three-layer CNN-based models are proposed to reconstruct the super-resolution images using four optimization algorithms, i.e., stochastic gradient descent (SGD), adaptive gradient (AdaGrad), root mean square prop (RMSprop), and adaptive moment estimation (ADAM). Among these four optimizations, ADAM is considered to have the best performance. To further verify the impact of the number of convolution layers on performance, a selection of CNN-based models with four convolutional layers is then proposed, each of which is named with the convolution parameters. All the four-layer models are optimized with ADAM, and experimental results indicate that the 9-3-3-5 model achieves the best performance in the super-resolution reconstruction task.

The existing methods of generating adversarial texts usually change the original meanings of texts significantly and even generate the unreadable texts. These less readable adversarial texts can misclassify the machine classifier successfully, but they cannot deceive the human observers very well. In this paper, we propose a novel method that generates readable adversarial texts with some perturbations that can also confuse human observers successfully. Based on the continuous bag-of-words (CBOW) model, the proposed method looks for the appropriate perturbations to generate the adversarial texts through controlling the perturbation direction vectors. Meanwhile, we apply adversarial training to regularize the classification model and extend it to semi-supervised tasks with virtual adversarial training. Experiments are conducted to show that the generated adversaries are interpretable and confused to humans and the virtual adversarial training effectively improves the robustness of the model.