Behavioral biometrics based continuous authentication has been proven to be an effective supplement to traditional one-time authentication schemes ( like passwords), and it can continuously authenticate users throughout the session. Currently, most continuous authentication models leverage deep learning techniques to learn smartphone users' behavioral patterns from behavioral biometric data, and have made remarkable progress. However, training deep learning based continuous authentication models requires enormous computing power, which leads to resource-constrained mobile platforms relying on powerful cloud servers. Cloud servers need access to behavioral biometric data for training and inference, as this mayraise privacy concerns. Existing deep learning based continuous authentication schemes pay more attention to authenticationperformance, but ignore theprotectionof behavioral biometric data. To solve this issue, we present a privacy-preserving continuous authentication scheme based on secret sharing secure multi-party computation (MPC). More specifically, we secretly share behavioral biometric data among two cloud servers that train continuous authentication systems on joint data using two-party computation (2 PC) without compromising data privacy. Further, we evaluate the practical feasibility of our proposed privacy-preserving scheme on two realistic privacy-preserving continuous authentication models, which are constructed with deep learning and traditional machine learning techniques, respectively. Extensive experiments demonstrate the effectiveness of our two privacy-preserving continuous authentication models.

As knowledge graphs become more diverse and complex, entity alignment tasks have been proposed to help integrate knowledge graphs from different sources. Existing methods for entity alignment are limited by the noise inherent in heterogeneous graphs, which hampers alignment performance. Additionally, most methods rely on entity names or other attributes for information propagation within the graph, but this kind of information often lacks sufficient semantics. To address these issues, we propose a novel model, Bert and Relation Graph Attention Network For Entity Alignment, called BRGEA. BRGEA employs a relation hierarchical graph attention network to mitigate noise propagation and leverage relation information, and BRGEA utilizes a pre-trained model to capture the long-text descriptions of entities, extracting richer semantic information. Experimental results on three real-world cross-language datasets show that our BRGEA model outperforms current advanced entity alignment methods.

In today’s data-driven era, data privacy protection is a key challenge in artificial intelligence development. Federated learning (FL) effectively addresses data silos and privacy concerns, but the statistical heterogeneity among clients limits the performance and generalization of traditional FL methods. Personalized federated learning (pFL) tackles this by training customized models for each client based on local data distributions. However, current pFL methods either focus on fine-tuning the global model on the client side or generating personalized models on the server side, failing to balance global collaboration and personalized learning. To address this, we propose ProtoFedLA, an innovative pFL framework that uses data prototypes to guide personalized model updates. This ensures consistency with the global data distribution while adapting to local data through localized aggregation. We evaluated ProtoFedLA using five publicly available image/text classification datasets under two heterogeneous data environments. The results show significant performance improvements over state-of-the-art pFL methods, with ProtoFedLA outperforming the best baseline by 5.83%. Additionally, ProtoFedLA handles data heterogeneity effectively, while demonstrating excellent stability and scalability.

With the development of telemedicine diagnosis technology, the collection, storage, and transmission of medical data has become a pivotal problem. To solve these problems, in terms of semi-tensor product compressed sensing (STP-CS) and hybrid cloud, a new medical data transmission framework is presented in this article, which can ensure the efficiency, confidentiality, and verifiability of data transmission. According to the edge detection, the authentication information is embedded into the insignificant area of the medical image to protect the authenticity of the data. STP-CS is utilized to measure and encrypt the medical image, which improves the efficiency of data transmission. In order to make the image data more secure, an adaptive cyclic shift diffusion algorithm based on chaotic sequence is used to effectively diffuse the measurement results. In addition, we also apply encoding for the quantized measurement value to be used as tamper proof authentication. The simulation results prove that the presented medical data transmission scheme can effectively improve the image reconstruction effect, transmission efficiency, and security.

The emergence of strong convective weather has caused serious threats to people’s lives and production activities, and also caused serious damage to the ecological environment. Because this kind of weather has a short duration and a small impact area, it is difficult to be accurate predicted. Doppler weather radar is one of the devices that can sample strong convective weather continuously for a long time, and plays an extremely important role in the detection and prediction of severe weather. The prediction of radar echo intensity by using deep learning methods has largely improved the prediction accuracy of strong convective weather, but the prediction of radar echo shape change is not yet ideal, so the trend of strong convective weather cannot be accurately judged. This paper proposes a weather radar echo prediction method based on cyclic convolutional neural network, which is called the EDD model, its design was inspired by the RU_Net families of deep learning models. The feature fusion layer was added to the neural network, thus ensuring that the feature matrix contains more feature parameters and improving the problem of inaccurate prediction of radar echo shape change due to insufficient information during upsampling. The EDD model is trained using Doppler weather radar echo maps, and the experimental results show that the EDD model used in this paper can improve the prediction accuracy of radar echo intensity and make good predictions of radar echo shape changes, thus helping meteorologists to make more accurate judgments on the occurrence, development and change trends of strong convective weather.

As a cornerstone service for many Internet of Things applications, channel state information (CSI)-based activity recognition has received immense attention over recent years. However, recognition performance of general approaches might significantly decrease when applying the trained model to the left-out user whose CSI data are not used for model training. To overcome this challenge, we propose a semi-supervised generative adversarial network (GAN) for CSI-based activity recognition (CsiGAN). Based on the general semi-supervised GANs, we mainly design three components for CsiGAN to meet the scenarios that unlabeled data from left-out users are very limited and enhance recognition performance: 1) we introduce a new complement generator, which can use limited unlabeled data to produce diverse fake samples for training a robust discriminator; 2) for the discriminator, we change the number of probability outputs from k + 1 into 2k + 1 (here, k is the number of categories), which can help to obtain the correct decision boundary for each category; and 3) based on the introduced generator, we propose a manifold regularization, which can stabilize the learning process. The experiments suggest that CsiGAN attains significant gains compared to the state-of-the-art methods.