As a typical application of edge intelligence, 3D object detection in autonomous driving often requires multimodal information fusion to accurately perceive the environment. With images and point clouds serving as critical sensory data sources, 3D object detection integrates multimodal fusion to enhance detection accuracy. Generally, fusion algorithms leveraging attention mechanism can intelligently extract and integrate multimodal sensing information to overcome limitations posed by sensor calibration. However, attention mechanism may cause challenges such as slow model convergence and high false positives. Therefore, in this paper, we propose the deformable denoising (DefDeN) model to effectively integrate modules including gated information fusion networks, multi-scale deformable attention mechanisms, noise addition and denoising method, and contrastive learning for multi-sensor feature fusion. Experimental results on the nuScenes dataset demonstrate the superiority of DefDeN in detection accuracy, and the effectiveness of precise and stable perception for complex scenarios in autonomous driving systems.

Federated Learning (FL) is an emerging distributed machine learning paradigm that provides privacy guarantees for training robust models on distributed clients. The primary challenge of FL is data heterogeneity, which slows down model convergence and degrades model performance. Knowledge distillation has recently demonstrated effectiveness in addressing this challenge. However, these approaches neglect the statistical heterogeneity in local models and the uncertainty of the data distribution in the global model, which results in the ensemble knowledge cannot be fully utilized to guide local model learning. In this work, we propose an unsupervised knowledge distillation method migrating the local class-level pseudo-data sample scheme in the server for fine-tuning the global model. Specifically, we provide the conditional autoencoder for each client to maintain a dynamic generator in the server, which ensembles the client’s class-level information. The proposal produces an auxiliary dataset representing the global class-level distribution to regulate the local model as an inductive knowledge bias, and employs unsupervised knowledge distillation to enhance the aggregated model’s performance. The extensive experiments show that our proposal significantly outperforms the current state-of-the-art FL algorithms and can be integrated as a flexible plugin into existing FL optimization algorithms to enhance model performance.

Accurate Photovoltaic (PV) generation forecasts can reduce power redeploy from the grid, thus increasing the supplier’s profit in the day-ahead electricity market. However, the PV process is affected differently by various factors under different weather conditions, resulting in significantly different energy output curves. In this context, this paper proposes a day-ahead PV power forecasting method with weather conditioned attention mechanism. We propose a Multi-Stream Attention Fusion Network (MSAFN) which utilizes an algorithm to derive the optimal decomposition algorithm for different weather conditions. The proposed Conditional Decomposition (CD) algorithm searches for the decomposition algorithms and corresponding hyperparameters of the prediction model, aiming to achieve the optimal prediction performance. The MSAFN incorporates multiple attention modules to learn the energy output patterns under various weather conditions. Notably, the attention modules adeptly learn patterns under diverse conditions, while simultaneously, the sharing of weights among the remaining components of the model effectively enhances prediction accuracy and facilitates a reduction in training time. We compare the state-of-the-art decomposition algorithms (VMD, EEMD, MSTL, etc.) and prediction models (BPN, LSTM, XGBoost, transformer, etc.) commonly used in PV prediction. The results show that the MSAFN model is more accurate than the models above, which has a noticeable improvement compared to other recent day-ahead PV predictions on Desert Knowledge Australia Solar Centre (DKASC) dataset.

Geological hazard risk assessment and severity prediction are of great significance for disaster prevention and mitigation. Traditional methods require a long time to evaluate and rely heavily on human experience. Therefore, based on the key factors affecting landslides, this paper designs a geological disaster prediction model based on Monte Carlo neural network (MCNN). Firstly, based on the weights of evidence method, a correlation analysis was conducted on common factors affecting landslides, and several key factors that have the greatest impact on landslide disasters, including geological lithology, slope gradient, slope type, and rainfall, were identified. Then, based on the monitoring data of Lanzhou City, 18 367 data records were collected and collated to form a dataset. Subsequently, these multiple key influencing factors were used as inputs to train and test the landslide disaster prediction model based on MCNN. After determining the hyperparameters of the model, the training and prediction capabilities of the model were evaluated. Through comparison with several other artificial intelligence models, it was found that the prediction accuracy of the model studied in this paper reached 89%, and the Macro-Precision, Macro-Recall, and Macro-F1 indicators were also higher than other models. The area under curve (AUC) index reached 0.8755, higher than the AUC value based on a single influencing factor in traditional methods. Overall, the method studied in this paper has strong predictive ability and can provide certain decision support for relevant departments.

The emergence of coordinated and consistent macro behavior among self-interested individuals competing for limited resources represents a central inquiry in comprehending market mechanisms and collective behavior. Traditional economics tackles this challenge through a mathematical and theoretical lens, assuming individuals are entirely rational and markets tend to stabilize through the price mechanism. Our paper addresses this issue from an econophysics standpoint, employing reinforcement learning to construct a multi-agent system modeled on minority games. Our study has undertaken a comparative analysis from both collective and individual perspectives, affirming the pivotal roles of reward feedback and individual memory in addressing the aforementioned challenge. Reward feedback serves as the guiding force for the evolution of collective behavior, propelling it towards an overall increase in rewards. Individuals, drawing insights from their own rewards through accumulated learning, gain information about the collective state and adjust their behavior accordingly. Furthermore, we apply information theory to present a formalized equation for the evolution of collective behavior. Our research supplements existing conclusions regarding the mechanisms of a free market and, at a micro level, unveils the dynamic evolution of individual behavior in synchronization with the collective.

Compared with traditional environments, the cloud environment exposes online services to additional vulnerabilities and threats of cyber attacks, and the cyber security of cloud platforms is becoming increasingly prominent. A piece of code, known as a Webshell, is usually uploaded to the target servers to achieve multiple attacks. Preventing Webshell attacks has become a hot spot in current research. Moreover, the traditional Webshell detectors are not built for the cloud, making it highly difficult to play a defensive role in the cloud environment. SmartEagleEye, a Webshell detection system based on deep learning that is successfully applied in various scenarios, is proposed in this paper. This system contains two important components: gray-box and neural network analyzers. The gray-box analyzer defines a series of rules and algorithms for extracting static and dynamic behaviors from the code to make the decision jointly. The neural network analyzer transforms suspicious code into Operation Code (OPCODE) sequences, turning the detection task into a classification problem. Comprehensive experiment results show that SmartEagleEye achieves an encouraging high detection rate and an acceptable false-positive rate, which indicate its capability to provide good protection for the cloud environment.

Sentence Boundary Disambiguation (SBD) is a preprocessing step for natural language processing. Segmenting text into sentences is essential for Deep Learning (DL) and pretraining language models. Tibetan punctuation marks may involve ambiguity about the sentences’ beginnings and endings. Hence, the ambiguous punctuation marks must be distinguished, and the sentence structure must be correctly encoded in language models. This study proposed a component-level Tibetan SBD approach based on the DL model. The models can reduce the error amplification caused by word segmentation and part-of-speech tagging. Although most SBD methods have only considered text on the left side of punctuation marks, this study considers the text on both sides. In this study, 465 669 Tibetan sentences are adopted, and a Bidirectional Long Short-Term Memory (Bi-LSTM) model is used to perform SBD. The experimental results show that the F1-score of the Bi-LSTM model reached 96 $\mathrm{\%}$, the most efficient among the six models. Experiments are performed on low-resource languages such as Turkish and Romanian, and high-resource languages such as English and German, to verify the models’ generalization.

Unpredictable and irreproducible digital keys are required to modulate security-related information in secure communication systems. True random number generators (TRNGs) rather than pseudorandom number generators (PRNGs) are required for the highest level of security. TRNG is a significant component in the digital security realm for extracting unpredictable binary bitstreams. Presently, most TRNGs extract high-quality "noise" from unpredictable physical random phenomena. Thus, these applications must be equipped with external hardware for collecting entropy and converting them into a random digital sequence. This study introduces a lightweight and efficient true random number generator (LETRNG) that uses the inherent randomness of a central processing unit (CPU) and an operating system (OS) as the source of entropy. We then utilize a lightweight post-processing method based on XOR and fair coin operation to generate an unbiased random binary sequence. Evaluations based on two famous test suites (NIST and ENT) show that LETRNG is perfectly capable of generating high-quality random numbers suitable for various GNU/Linux systems.

3D object detection is a critical technology in many applications, and among the various detection methods, pointcloud-based methods have been the most popular research topic in recent years. Since Graph Neural Network (GNN) is considered to be effective in dealing with pointclouds, in this work, we combined it with the attention mechanism and proposed a 3D object detection method named PointGAT. Our proposed PointGAT outperforms previous approaches on the KITTI test dataset. Experiments in real campus scenarios also demonstrate the potential of our method for further applications.

With the development of automobile intelligence and connectivity, Intelligent and Connected Vehicle (ICV) is an inevitable trend in the transformation and upgrading of the automotive industry. The maturity of any advanced technology is inseparable from a large number of test verifications, especially the research and application of automotive technology require a large number of reliable tests for evaluation and confirmation. Therefore, the ICV Test Site (ICVTS) will become a key deployment area. In this paper, we analyze the development status of ICVTS outside and within China, summarize the shortcomings of the existing test sites, and put forward some targeted suggestions, in an effort to guide the development and construction of ICVTS towards the path that seems to be most promising.