In recent years, artificial intelligence technologies have been widely used in computer vision, natural language processing, automatic driving, and other fields. However, artificial intelligence systems are vulnerable to adversarial attacks, which limit the applications of artificial intelligence (AI) technologies in key security fields. Therefore, improving the robustness of AI systems against adversarial attacks has played an increasingly important role in the further development of AI. This paper aims to comprehensively summarize the latest research progress on adversarial attack and defense technologies in deep learning. According to the target model’s different stages where the adversarial attack occurred, this paper expounds the adversarial attack methods in the training stage and testing stage respectively. Then, we sort out the applications of adversarial attack technologies in computer vision, natural language processing, cyberspace security, and the physical world. Finally, we describe the existing adversarial defense methods respectively in three main categories, i.e., modifying data, modifying models and using auxiliary tools.

Physics-informed neural networks (PINNs) have emerged as a transformative methodology integrating deep learning with scientific computing. This review establishes a three-dimensional analytical framework to systematically decode PINNs’ development through methodological innovation, theoretical breakthroughs, and cross-disciplinary convergence. The contributions include threefold: First, identifying the co-evolutionary path of algorithmic architectures from adaptive optimization (neural tangent kernel-guided weighting achieving 230% convergence acceleration in Navier-Stokes solutions) to hybrid numerical-deep learning integration (5× speedup via domain decomposition) and second, constructing bidirectional theory-application mappings where convergence analysis (operator approximation theory) and generalization guarantees (Bayesian-physical hybrid frameworks) directly inform engineering implementations, as validated by $72\%$ cost reduction compared to FEM in high-dimensional spaces ($p<0.01,n=15$ benchmarks). Third, pioneering cross-domain knowledge transfer through application-specific architectures: TFE-PINN for turbulent flows ($5.12\pm 0.87\%$ error in NASA hypersonic tests), ReconPINN for medical imaging ($▵\mathrm{SSIM}=+0.18\pm 0.04$ on multi-institutional MRI), and SeisPINN for seismic systems ($0.52\pm 0.18$ km localization accuracy). We further present a technological roadmap highlighting three critical directions for PINN 2.0: neuro-symbolic, federated physics learning, and quantum-accelerated optimization. This work provides methodological guidelines and theoretical foundations for next-generation scientific machine learning systems.

In recent years, due to the frequent occurrence of network intrusions, more and more researchers have begun to focus on network intrusion detection. However, it is still a challenge to detect unknown attacks. Currently, there are two main methods of unknown attack detection: clustering and honeypot. But they still have unsolved problems such as difficulty in collecting unknown attack samples and failure to detect on time. Zero-Shot learning is proposed to deal with the problem in this article, which can recognize unknown attacks by learning the mapping relations between feature space and semantic space (such as attribute space). When the semantic descriptions of all attacks (including known and unknown attacks) are provided, the classifier built by Zero-Shot learning can extract common semantic information among all attacks and construct connections between known and unknown attacks. The classifier then utilizes the connections to classify unknown attacks although there are no samples for unknown attacks. In this article, we first propose to use Zero-Shot learning to overcome the challenge of unknown attack detection and illustrate the feasibility of this method. Secondly, we then propose a novel method of Zero-Shot learning based on sparse autoencoder for unknown attack detection. This method maps the feature of known attacks to the semantic space, and restores the semantic space to the feature space by constrains of reconstruction error, and establishes the feature to semantic mapping, which is used to detect unknown attacks. Verification tests have been carried out by using the public dataset NSL_KDD. From the experiments conducted in this work, the results show that the average accuracy reaches 88.3%, which performs better than other methods.

In recent years, the single-modal spam filtering systems have had a high detection rate for image spamming or text spamming. To avoid detection based on the single-modal spam filtering systems, spammers inject junk information into the multi-modality part of an email and combine them to reduce the recognition rate of the single-modal spam filtering systems, thereby implementing the purpose of evading detection. In view of this situation, a new model called multi-modal architecture based on model fusion (MMA-MF) is proposed, which use a multi-modal fusion method to ensure it could effectively filter spam whether it is hidden in the text or in the image. The model fuses a Convolutional Neural Network (CNN) model and a Long Short-Term Memory (LSTM) model to filter spam. Using the LSTM model and the CNN model to process the text and image parts of an email separately to obtain two classification probability values, then the two classification probability values are incorporated into a fusion model to identify whether the email is spam or not. For the hyperparameters of the MMA-MF model, we use a grid search optimization method to get the most suitable hyperparameters for it, and employ a k-fold cross-validation method to evaluate the performance of this model. Our experimental results show that this model is superior to the traditional spam filtering systems and can achieve accuracies in the range of 92.64–98.48%.