The rapid development of blockchain technology and the rise of cryptocurrencies not only challenge traditional financial and asset management concepts but also introduce significant privacy and security concerns. Coin mixing, a primary privacy protection mechanism in blockchains, is designed to enhance transaction anonymity and protect user identities. However, this enhanced anonymity also enables the misuse of cryptocurrencies for illicit activities such as money laundering, theft, and fraud. Therefore, a comprehensive review of technologies for the identification, detection, and regulation of coin mixing services is urgently required to support law enforcement and promote the secure and compliant use of cryptocurrencies. This paper first reviews the technical principles and mechanisms of coin mixing services and proposes a general classification framework for mixing tools while discussing the criminal activities associated with them. Subsequently, it focuses on recent advances in five key regulatory approaches: manual rule-based methods, address clustering, entity classification, anomalous transaction detection, and mixing service traceability. It also surveys existing industrial tools and platforms related to coin mixing regulations. Finally, it identifies current challenges in the field and suggests directions for future research.

Deep learning has achieved significant success in computer vision and speech signal processing. However, its rapid development of deep learning has had negative effects. Many types of fake videos and voices are flooding the Internet. Some criminals use deep learning technology to replace the face in original video, edit facial attributes, and synthesize or clone the speaker's voice. Criminals can cause social unrest and chaos by producing pornographic videos, fake news, and political rumors, which threaten personal interests and national security. Many scholars have proposed solutions to eliminate these negative effects. Early forgeries focused primarily on single-modal forgeries. Therefore, most current solutions focus on single-modal counterfeit detection and fail to fully consider the intrinsic relationship between audio and video. Existing single-modal detection methods often exhibit suboptimal performance when both audio and video are forged. Recently, with the increase in research, some scholars have begun to explore the use of multimodal models for counterfeit detection and have achieved remarkable results. This survey reviews video and voice forgery and detection technologies, collects and sorts voice, audio, and video forgery datasets, and summarizes multimodal counterfeit detection methods. Finally, the existing problems and future research directions of the current detection technology are analyzed.

Time series anomaly detection is a critical task in the finance, medical treatment, and industrial monitoring fields. However, when handling time series data, traditional methods often face challenges such as limited feature extraction ability, poor generalization ability, and poor real-time performance. These methods may even ignore the temporal relationships between data. To fully consider the temporal dependencies in time series data and further enhance the ability to detect anomalies, this paper proposes an unsupervised time series anomaly detection method called β-VAE-BiLSTM. First, the Variational Autoencoder (VAE) network structure is customized for data reconstruction in anomaly detection. Next, a hyperparameter, β, is introduced to control the weight of the Kullback-Leibler (KL) divergence term in the evidence lower bound, enhancing the encoder module's disentanglement and expressive capabilities in the latent space and obtaining more robust data representations. Then, Bidirectional Long Short-Term Memory (BiLSTM) is used to estimate the long-term correlations of the latent representations, capturing their forward and backward dependencies and performing temporal predictions. Finally, the reconstruction error is calculated by averaging the fusion of the prediction results and the decoder reconstruction. The optimal threshold for anomaly detection is determined through grid search. Experimental results show that the proposed method has superior anomaly detection performance on multiple public time series datasets. It can effectively extract complex temporal data features, achieve efficient computation and real-time anomaly detection, identify anomalous points and regions, improve detection accuracy and robustness, and reduce false positives and false negatives.

The Transformer model, owing to its excellent global information capture and powerful representation abilities, has been widely used in medical image segmentation and has achieved remarkable results. However, when serializing images, these methods divide them into fixed-sized blocks and extract only global features of a single scale, which, to some extent, fragments the semantic features of the images, ultimately leading to poor segmentation accuracy. To address this issue, this paper proposes a Multi-scale self-attention Transformer architecture (MultiFormer). First, continuous convolution and downsampling modules are used to process images. Then, the original 1×1 projection module is replaced with a multi-scale convolution projection module. Finally, deformable convolution is introduced into the feature maps generated by the self-attention module. Compared to the traditional Transformer image serialization process, this continuous convolution effectively enlarges the receptive field while generating features of the same resolution, retains the spatial correlation of 2-Dimensional (2D) images, and avoids the loss of semantic information caused by fixed-position and fixed-size image blocks. Meanwhile, the multi-scale convolution projection module captures contextual information in the images using four convolution kernels of different sizes and achieves multi-scale feature fusion through channel concatenation, reflecting attention between local regions at different scales, rather than being limited to a single scale, which enables the model to aggregate semantic information at different scales and further alleviates the problem of semantic fragmentation. Deformable convolution introduces an additional convolution layer to learn and generate an offset field, allowing the convolution kernel to flexibly adjust its shape to adapt to morphologically diverse lesions or organs in the images and enhancing the ability of the model to process complex medical images. This module is inserted into three network structures—SETR, TransUNet, and TransFuse—and experiments are conducted on the ACDC heart and ISIC 2018 skin lesion datasets. The results show that the Dice coefficient improves by 3.63, 1.06, and 2.30 percentage points and by 1.22, 2.31, and 3.01 percentage points, respectively. MultiFormer is a plug-and-play tool and can be easily integrated into various downstream medical image analysis tasks.

Structured medical data are widely used in real-world healthcare systems. Owing to their unique characteristics, they typically exhibit properties such as being multi-source and heterogeneous with many potential candidate features, resulting in irregular data patterns that pose challenges to the feature-processing procedures of neural networks. Therefore, a recurrent directional feature processing network with a feature-embedding level similar to that of the Transformer architecture, named RDFP-NET, is proposed. The network possesses a favorable feature organization pattern and performs end-to-end processing directly on the collected raw medical data, enabling it to fully leverage effective information while eliminating the negative impact of unfavorable features. In addition, it employs a recurrent structure to correlate contextual feature information, further enhancing network task performance. RDFP-NET mainly comprises: a nonlinear feature-embedding method; a ranked feature interaction module that incorporates inductive bias into the general attention mechanism by calculating the information value of each feature and applying specific masking operations, achieving directional information flow and feature interaction among features with varying contribution values; a filtered feature selection module that discards redundant features in a filter-like manner by adjusting the Exponential Linear Unit (ELU) activation function; and a recurrent structure comprising two modules that facilitates good data fitting and contextual feature information flow while gradually addressing irregular feature patterns. Experiments conducted on five datasets encompassing two different target tasks demonstrate that RDFP-NET achieves an optimal average rank of 2.2 and a rank standard deviation of 0.75, showing excellent task performance and stability. Thus, RDFP-NET can provide medical professionals with more effective auxiliary information for clinical decision-making.

World models are generally believed to understand and represent the external world and predict future states based on current world states and actions. Large models leverage massive training data and vast parameter scales to exhibit outstanding capabilities in learning, understanding, representing, and generating textual knowledge, as exemplified by language large models such as GPT-4 and LLaMA. In recent years, research on world models has attracted significant attention from both industry and academia, leading to significant research and commercial achievements in domains such as autonomous driving, social simulation, embodied intelligence, and video generation. Moreover, researchers have applied the remarkable results of various large models to world models, further enhancing their performance. This paper comprehensively reviews world models built using large models across different domains, covering both language large model- and Vision Large Model (VLM)-based approaches. Several important application areas, including embodied intelligence, smart cities, social simulation, and physical environment simulation, are selected to introduce relevant models. This paper classifies world models based on the modality of the large models used, highlighting the functional differences between world models based on different modalities. Subsequently, important open-source resources and benchmarks for world models are presented to help researchers in related fields understand and utilize world models quickly. Finally, this paper is summarized and future research directions are presented.

Multimodal sentiment recognition aims to improve the accuracy and robustness of sentiment detection by integrating information from different modalities such as text, audio, and video. However, existing methods face challenges in handling discrepancies and complementarities between modalities, as well as in capturing the dynamic features of temporal sequences, often resulting in suboptimal sentiment recognition performance. To address these issues, this paper proposes a multimodal sentiment recognition model based on cross-modal enhancement and a time-step gating mechanism. The model employs a cross-modal cross-attention mechanism to learn correlations between different modalities, thereby enhancing the complementarity of features across modalities. The model integrates information from text, audio, and video through interactions between modalities, mitigating the limitations of single-modality sentiment expressions. Subsequently, the time-step gating mechanism dynamically adjusts feature weights at each time-step, focusing on critical time-steps that contain more relevant sentiment information, thereby improving the model's temporal sequence modeling capability. Finally, fused features are fed into a classifier for sentiment prediction. Experimental evaluations on publicly available CMU-MOSEI and CMU-MOSI multimodal sentiment recognition datasets show that the proposed model achieves sentiment recognition accuracies of 82.41% and 82.60%, respectively, significantly outperforming current mainstream models such as ALMT and TETFN. These results demonstrate that cross-modal enhancement and time-step gating mechanisms effectively improve the ability to fuse multimodal features and process temporal sequences, validating the effectiveness and robustness of the method in multimodal sentiment recognition tasks.

To address the issue of interference factors such as seasonal changes, climate, and illumination that affect high-resolution remote sensing images of the same geographical space but different temporal phases, a remote sensing image building Change Detection (CD) method based on a multi-temporal ChangeFormer is proposed. This method uses multiple remote sensing images from different temporal phases and fuses the latest temporal remote sensing image with multiple prechange remote sensing images at different scales for feature difference extraction. Additionally, it focuses on both the comprehensive semantic features of the images and the details of the semantic information between the images. This approach helps reduce false detections caused by changes in factors such as season and illumination. Additionally, the method fuses the feature differences of multiple prechange remote sensing images from different temporal phases and introduces a regularization term as a loss function. This eliminates interference from nonbuilding changes and illumination shadows in the nonchanging areas of buildings, thereby enhancing the generalization ability of the model. A three-temporal remote sensing image dataset covering changes from agricultural land to construction land is constructed. The experimental results show that, compared to the current optimal BIT method, the multi-temporal ChangeFormer method improves the F1 value, Intersection over Union (IoU), precision, and recall by 9.04%, 9.87%, 15.27%, and 3.4%, respectively, thus significantly enhancing detection accuracy. Furthermore, it outperforms classical CD methods in terms of detailed information processing.

In Printed Circuit Board Assembly (PCBA), defect detection is key to improving production line efficiency. However, after assembly, printed circuit boards are usually inspected manually, leading to labor and time wastage, as well as missed and false detections. To address these issues, this paper proposes an improved lightweight YOLOv8s network that effectively reduces model complexity while enhancing the accuracy of PCBA defect detection. First, owing to the lack of publicly available PCBA-related datasets, a dataset called PCBA-DET is constructed for post-assembly PCBA defect detection. Various data augmentation techniques are applied to PCBA-DET to simulate real-world production scenarios and improve the dataset balance. Second, the last C2f module of the YOLOv8s backbone is replaced with a Re-parameterized Large Kernel convolution Network (RepLKNet) to reduce computational cost and increase the effective receptive field of the model. In addition, in the neck network of YOLOv8s, a P2 small object detection layer and Ghost Convolution are introduced to capture more detailed information and effectively reduce the number of model parameters. On the augmented PCBA-DET dataset, the improved model achieves an increase of 2.6 and 0.1 percentage points in terms of mean Average Precision (mAP)@0.5∶0.95 and mAP@0.5, respectively, compared with the baseline model, whereas the number of parameters is reduced by 36.8%.

When convolutional networks are used in the image inpainting of cultural relics, the convolution kernel's limited receptive field poses challenges, which results in a weak comprehension of the global context and complex structures. Moreover, the convolution operation does not adequately handle the intricate geometrical shapes of relic surfaces owing to its translation invariance; hence, convolution-based inpainting is prone to irrelevant structures and artifacts. In the case of Transformer models with self-attention mechanisms, which process the details and local features of relic images, the insufficient attention to specific regions makes it difficult to capture the deep features necessary for precise and detailed inpainting. Additionally, Transformers cannot adequately capture long-range semantics, which results in a suboptimal visual quality of the inpainted images. This paper proposes a relic image inpainting model based on the SwinTransformer, called the Dynamic Mask on SwinTransformer (DMSWT). The model introduces several improvements to the self-attention module within the network to optimize its structure. First, layer normalization is removed, and fully connected layers are replaced with residual connections to enhance the deep feature extraction capabilities of the network. Second, a dynamic mask mechanism is introduced to mitigate the issue of reduced effective pixels caused by default sampling in the inpainting of images with large-scale missing regions. Finally, the loss function is improved with a focus on enhancing the perceptual realism, leading to an improvement in the visual quality of the inpainted images. Experimental results for different scenarios show that the DMSWT model can learn more structural prior information and generate inpainted images that align with real-world intuition. Additionally, quantitative evaluations demonstrate significant improvements in performance metrics.

Timely and effective disinformation detection is crucial for curbing the spread of disinformation and minimizing social harm. Numerous deep learning methods have been employed for disinformation detection. Summarizing the detection principles and paradigms of existing research is essential for identifying directions for technical optimization. Therefore, this paper comprehensively reviews existing research based on the principles and implementation paths of disinformation detection, and for the first time, summarizes and compares the applications of large language models in this field. First, the relevant concepts of disinformation detection tasks are introduced and the data structures of commonly used disinformation detection datasets are summarized. Then, based on detection principles and implementation methods, the paper presents ways to detect textual and multimodal disinformation through semantic feature representation, auxiliary task design, internal knowledge inference, and fact verification, refining them into ten subcategories and summarizing the potential characteristics of detection methods for each subcategory. Finally, the paper summarizes disinformation detection paradigms based on deep neural networks and large language models, compares the detection performance of representative methods from these paradigms across seven disinformation detection datasets, and highlights the advantages and limitations of large language models in detecting disinformation. It also presents the anticipated opportunities and challenges brought about by large language models in the field of disinformation detection, providing a reference for future research.

In multi-agent game simulations, the performance of Large Language Model (LLM) has been widely studied; however, their decision-making ability to guide multi-agent cooperation in fuzzy task objectives or uncertain environments is often unreliable. To address this issue, a multi-level collaborative decision-making framework based on distributed Bayesian inferences is proposed. This framework integrates three major functional modules: decision making, peer evaluation, and supervision. It utilizes multiple LLM for collaborative decision making and has been experimentally validated in a spatial prisoner's dilemma game. The experimental results show that the framework effectively overcomes the decision-making bottleneck of LLM in fuzzy task environments and successfully promotes the emergence of multi-agent cooperative behavior. Additionally, a quantitative evaluation of the model's decision-making ability in different experimental scenarios reveals that the decision error of the model is not linearly related to the model size. Under fuzzy task instructions, the decision error of the LLaMA3 (70×10⁹) model is 16.6% higher than that of the LLaMA3 (8×10⁹) model and 7.2% higher than that of the LLaMA2 (7×10⁹) model. This indicates that in more complex environments, relying solely on the expansion of the model size does not significantly improve the decision-making performance. By contrast, LLM collaborative decision making has shown significant advantages in improving decision consistency and effectiveness. These results reveal the crucial role of multi-model collaboration in complex decision-making environments and provide important references for the future design of intelligent agent systems for uncertain tasks.

Rolling bearings are components commonly used in mechanical equipment. Traditional methods struggle to classify signals with numerous complex features in a multi-noise environment. They often rely on classical deep learning models for performing fault diagnosis using one-dimensional data, failing to fully extract complex features. To address this issue, this paper proposes a dual-channel fault diagnosis method based on the ACNN-LFSwin Transformer, which performs fault diagnosis on both one-dimensional data and two-dimensional images. First, the original signal is processed using Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Short-Time Fourier Transform (STFT) to obtain Intrinsic Mode Functions (IMF) and two-dimensional images. Subsequently, in channel 1, the CEEMDAN-decomposed IMF are fed into an Attention-based Convolutional Neural Network (ACNN) for feature extraction. In channel 2, the two-dimensional images composed of bearing data are input into a Swin Transformer network (LFSwin Transformer) for local feature extraction. Finally, the features from both channels are concatenated and fused for fault diagnosis. ACNN employs an attention mechanism to automatically allocate weights to signal features, thereby emphasizing key features. The LFSwin Transformer performs vector conversion based on the traditional Swin Transformer, converts the input vector into an image, and performs convolution operations, making the model more advantageous in extracting local fault features. In experiments on the CWRU and Paderborn datasets, the proposed method achieves a fault diagnosis accuracy of over 97%. This result shows that it can accurately diagnose various faults and effectively avoid interference from complex noise.

Automatic Modulation Recognition (AMR) is a crucial component in communication identification, situational awareness, and electronic reconnaissance. Deep neural networks, known for their powerful feature extraction and classification capabilities, offer higher recognition accuracy compared to traditional methods. However, current neural networks exhibit limitations in effectively extracting temporal information from signals, leading to high complexity and poor recognition accuracy under low Signal-to-Noise Ratio (SNR) conditions. To address these issues, this paper proposes a decision fusion recognition scheme based on a Residual Neural Network (ResNet) and Transformer network (ResNet-Transformer). This scheme aims to handle more complex SNR scenarios and improve the overall recognition accuracy. By leveraging the temporal memory characteristics of ResNet to deeply extract time-domain features from communication signals, and combining the outstanding long-distance dependency extraction capabilities of the Transformer network to enhance noise resistance, the proposed scheme employs a decision fusion strategy to obtain the final decision based on the outputs of each branch. Experimental results show that the proposed scheme achieves an average recognition accuracy of over 93% for SNRs above 10 dB and maintains a recognition accuracy of 56% even at an SNR of 0 on the open dataset RML2018.01A. Compared to traditional network models, the proposed scheme achieves a higher modulation recognition accuracy and exhibits a high noise resistance.

Remote areas face problems such as insufficient cellular network coverage as well as low energy and computing power of Internet of Things (IoT) devices. Hence, the requirements for delay-sensitive task offloading and computing for a large number of tasks cannot be met. Considering the combination of the Space—Air—Ground Integrated Network (SAGIN) and Mobile Edge Computing (MEC), this paper proposes a strategy for dynamic task offloading and resource allocation for Unmanned Aerial Vehicle (UAV)-assisted IoT devices that support Wireless Power Transmission (WPT) technology, in which UAVs are responsible for collecting compute-intensive tasks generated by IoT devices. These tasks are locally calculated or dynamically unloaded to the base station and a Low Earth Orbit (LEO) satellite for further processing using a partial unloading mode, according to the current state. Given the dynamic heterogeneous network environment, as well as the tight coupling between long-term queuing delays and short-term decision-making, this paper proposes a Twin Delayed Deep Deterministic Policy Gradient (TD3PG) algorithm based on Lyapunov optimization under queuing delay constraints. The algorithm coordinates UAVs to learn the optimal offloading strategy and resource allocation by optimizing UAV dynamic association, task allocation, computing resource allocation, and bandwidth allocation. Simulation results show that, compared with other schemes, the proposed dynamic scheme can effectively reduce the energy consumption, network backlog sum, and average queue delay in the UAV network. Under different learning rate combinations, the reward of the TD3PG algorithm increases by 13.6% and 24.0% compared with that of the Deep Deterministic Policy Gradient (DDPG) algorithm, and by 20.4% and 17.9% compared with that of the Double Deep Q-Network (DDQN) algorithm.

Traditional Retrieval-Augmented Generation (RAG) methods predominantly focus on pure-text scenarios. In these scenarios, their retrieval and generation mechanisms encounter difficulties in effectively modeling common visual elements, spatial layouts, and structural semantics within multimodal documents. This drawback restricts their performance in tasks related to text-image hybridization, long documents, and cross-document reasoning. To tackle this issue, Multimodal Retrieval Augmented Generation (MRAG), by integrating text, image, and layout structure modeling, and incorporating multimodal evidence retrieval and scheduling during the generation process, has already developed into a core technical paradigm for Question & Answer (Q & A) and reasoning in visually-rich documents. This paper conducts a systematic review of research progress in MRAG applications for document Q & A tasks. Firstly, based on the practical requirements for multimodal document understanding, we analyze the key challenges in MRAG implementation, including multimodal alignment, long-context modeling, evidence traceability, and system robustness. Secondly, from the perspective of how MRAG systems support the generation process, we compare representative methods across four dimensions: embedding paradigms, document retrieval scope, layout-aware mechanisms, and multimodal retrieval strategies. We focus on how design choices influence generation stability, reasoning accuracy, and system complexity. Thirdly, we summarize the characteristics and limitations of existing multimodal document Q & A datasets and evaluation frameworks, and analyze the current constraints in evidence granularity and reasoning explainability. Finally, we point out that MRAG is evolving from static similarity-matching retrieval mechanisms to dynamic evidence planning paradigms centered on generation and reasoning needs, and should continuously enhance the reliability and explainability of complex document Q & A systems through collaborative multimodal modeling with multi-granularity approaches.

Convolutional Neural Networks (CNNs) are widely used in the field of object detection, earning widespread acclaim in scholarly circles due to their precision and scalability. It has spawned numerous notable models, including those in the Region-based Convolutional Neural Networks (R-CNNs) (such as Fast R-CNN and Faster R-CNN) and You Only Look Once (YOLO) series. After the success of Transformers in the field of natural language processing, researchers began exploring their application in computer vision, leading to the development of visual backbone networks such as Visual Transformer (ViT) and Swin Transformer. In 2020, a Facebook research team unveiled DEtection TRansformer (DETR), an end-to-end object detection algorithm based on Transformers, designed to minimize the need for prior knowledge and postprocessing in object detection tasks. Despite the promise shown by DETR in object detection, it has limitations including low convergence speed, relatively low accuracy, and the ambiguous physical significance of target queries. These issues have spurred a wave of research aimed at refining and enhancing the algorithm. This paper aims to collate, scrutinize, and synthesize the various efforts aimed at improving DETR, assessing their respective merits and demerits. Furthermore, it presents a comprehensive overview of state-of-the-art research and specialized application domains that employ DETR and concludes with a prospective analysis of the future role of DETR in the field of computer vision.

The RSD-YOLO algorithm, based on YOLOv8s, is proposed to address the challenges of low detection performance, severe occlusion, difficulty of small target feature extraction, and large number of model parameters inherent in Unmanned Aerial Vehicle (UAV) aerial images. First, the Receptive Field Attention (RFA) module CSP-RFA is designed to replace the C2f module for enhancing the capability of small target feature extraction, effectively addressing the insensitivity of traditional convolutional operations to positional changes. Second, the backbone and feature fusion networks are made lightweight, a new large-size feature map detection branch is added, and a Receptive Field Pyramid Network (RFPN) is proposed to optimize the feature flow direction and improve feature representation. Third, the detection head module is optimized by integrating multi-scale features with a multi-level attention mechanism and the loss function is updated to improve the model's detection performance for small targets. Finally, in terms of model compression, Layer-Adaptive Magnitude-based Pruning (LAMP) algorithm is employed to further reduce the number of parameters and model size. The experimental results demonstrate that the lightweight RSD-YOLO model significantly outperforms the baseline model on the publicly available VisDrone2019 dataset, with a 10.0 percentage point increase in precision, a 9.5 percentage point increase in mAP@0.5 (equivalent to a 24.1% increase), and a 6.9 percentage point increase in mAP@0.5∶0.95 (equivalent to a 29.4% increase). The number of model parameters is reduced from 11.12×10⁶ to 4.05×10⁶, representing a 63.6% reduction, and the computational cost is reduced from 42.7 GFLOPs to 25.5 GFLOPs, showing a 40% reduction. Furthermore, for a newly filtered dataset focusing on small occluded targets, RSD-YOLO shows improvements of 9.1, 16.1, and 10.7 percentage points in terms of precision, mAP@0.5, and mAP@0.5∶0.95, respectively.

Federated learning enhances data sharing and collaboration between healthcare institutions, thereby improving the accuracy and efficiency of medical diagnoses, treatments, and predictions. However, existing federated learning solutions face security and efficiency challenges. Model parameter updates during training may inadvertently disclose information about local training datasets. To ensure parameter confidentiality, researchers have proposed various solutions such as masking protocols and differential privacy. However, masking protocols often lack strong security, whereas differential privacy leads to tradeoffs between accuracy and privacy. To address these challenges, this study proposes a secure federated learning scheme for smart healthcare based on secret sharing and homomorphic encryption. This scheme effectively prevents both healthcare clouds and clients from stealing model parameters and resists collusion attacks among participants. In addition, a ciphertext verification algorithm is used to ensure that model parameters can be verified during training. Security and performance analyses demonstrate that our scheme meets the confidentiality and integrity requirements for model parameters in smart healthcare scenarios, with significant improvements in computational and transmission efficiency compared to existing solutions.

Federated Learning (FL), a distributed machine learning technology, has achieved significant results in privacy protection. However, in practical applications, client drift phenomena occur because of the Non-Independent and Identically Distributed (Non-IID) nature of data sources, leading to slow model convergence and performance degradation. To address this issue, this study proposes a Federated Local Momentum accelerated learning (FedLM) algorithm combined with the attention mechanism. FedLM introduces a global momentum term into local model updates, utilizing the global gradient information from previous rounds to smooth the current update process and correct the divergence of parameter update directions among heterogeneous clients, thereby reducing gradient oscillations and alleviating data heterogeneity issues. The attention mechanism dynamically adjusts the weight of each client in the global model update to improve the quality of the aggregation model. Experimental results show that FedLM achieves significantly better accuracy and stability than existing federated learning algorithms such as SCAFFOLD, FedCM, and Moon in image classification tasks with different levels of data heterogeneity, model structures, and datasets.

With advances in communication technology, the Internet of Things (IoT) has played an increasingly important role in real life, and the application of Unmanned Aerial Vehicle (UAV) communication in the IoT has been widely studied. UAVs are used as mobile data collectors to collect data from Sensor Nodes (SNs) in a WSN. The Age of Information (AoI) is introduced as an index for evaluating network performance. This paper proposes a data collection scheme based on UAV trajectory design and a scheduling strategy for SNs. Based on this scheme, a weighted minimization model is constructed to minimize the weighted sum of the Average AoI (AAoI) and energy consumption of the SNs by optimizing the trajectory of the UAV, the scheduling of the SNs, and the transmitting power. This mixed-integer nonlinear problem is usually difficult to solve directly. Therefore, the path discretization method is first used to discretize multiple continuous variables. Subsequently, a joint optimization algorithm based on Block Coordinate Descent (BCD) and Successive Convex Approximation (SCA) is proposed to obtain a local optimal solution that satisfies the KKT condition. The simulation results show an effective balance between the AoI and the energy consumption of the SNs, demonstrating the feasibility of the proposed scheme.

Crowd intelligence has been highlighted as one of the five major trend directions in China's New Generation Artificial Intelligence Development Plan, and the question ″How does crowd intelligence emerge?″ was listed by Science in 2021 among 125 key scientific challenges for the future. Existing studies on intelligence emergence have largely focused on collective intelligence in biological groups (e.g., flocks of birds and schools of fish), emphasizing how simple individuals following local interaction rules can give rise to globally ordered behaviors. By contrast, crowd intelligence in human societies involves not only behavioral coordination and organization but also higher-order intelligence and richer connotations manifested in knowledge, culture, and innovation. To address this discrepancy and research gap, this study, for the first time, systematically reviews and clarifies the conceptual framework and core connotations of crowd intelligence in human societies. It conducts an in-depth analysis of seven representative emergent phenomena—crowd behavior, wisdom of crowds, consensus formation, social cooperation, social learning, knowledge and culture, and collective intelligence innovation—by summarizing their key mechanisms, major models, driving factors, and evolutionary regularities, thereby constructing a theoretical system for the emergence of crowd intelligence in human societies. Building on this foundation, the study further investigates the mapping pathways and mechanisms from human crowd intelligence to artificial crowd intelligence; extracts representative paradigms and essential implementation essentials of mechanism-driven artificial crowd intelligence systems; and provides fundamental theoretical and methodological support for the design, construction, and future development of artificial crowd intelligence systems oriented toward complex tasks.

Artificial intelligence has made remarkable progress across many fields, encouraging countries to attach great importance to its research and development. However, the rapid development of artificial intelligence has also brought about a series of problems and threats, and overreliance on and blind trust in such models can lead to serious risks. Therefore, interpretable artificial intelligence has become a key element in building trusted and transparent intelligent systems, and its research and development requires immediate attention. This survey comprehensively summarizes the research progress on explainable artificial intelligence at home and abroad comprehensively from multiple dimensions and levels. Based on current research results in the industry, this survey subdivides the key technologies of explainable artificial intelligence into four categories: interpretation model, interpretation method, safety testing, and experimental verification, with the aim of clarifying the technical focus and development direction of each field. Furthermore, the survey explores specific application examples of explainable artificial intelligence across key industry sectors, including but not limited to education, healthcare, finance, autonomous driving, and justice, demonstrating the significant role it plays in enhancing decision-making transparency. Finally, this survey provides an in-depth analysis of the major technical challenges of interpretable artificial intelligence and presents future development trends, in addition to a special investigation and in-depth analysis of the interpretability of large models, which has attracted considerable attention recently.

In view of missed and false detection phenomena caused by numerous small target instances and occlusions among targets in drone images, this paper proposes a lightweight small target detection algorithm for Unmanned Aerial Vehicle (UAV) images based on an improved YOLOv8. The Triple Feature Encoder (TFE) and Scale Sequence Feature Fusion (SSFF) modules are introduced in the neck to enhance the ability of the network to extract features at different scales. Furthermore, a Small Object Detection Head (SMOH) is designed and fused with the improved neck feature extraction network, and an additional detection head is also introduced to reduce the loss of small target features and enhance the recognition ability of the network for small targets. Additionally, considering the defects of Complete Intersection over Union (CIoU), a regression loss function, Wise-Inner-MPDIoU, is proposed by combining Wise-IoU, Inner-IoU, and Minimum Point Distance based IoU (MPDIoU). Finally, to realize the lightweight application requirements of the algorithm in mobile and embedded systems, amplitude-based layer-adaptive sparse pruning is performed to further reduce the model size while ensuring model accuracy. Experimental results demonstrate that, compared to the original YOLOv8s model, the improved model proposed in this paper improves mAP@0.5 by 6.8 percentage points, while reducing the number of parameters, amount of computation, and model size by 76.4%, 17.1%, and 73.5%, respectively. The proposed model is lightweight, improves detection accuracy, and has strong practical significance.

Owing to the heterogeneity and complexity of Android malware, traditional static analysis methods that rely on single features such as permissions or API often struggle to accurately differentiate between benign and malicious applications. To address this limitation, this study proposes a novel feature construction method based on multi-modal feature fusion based on in-depth research of Android software features such as permissions, API, bytecodes, and opcodes. The bytecode is transformed into RGB images and visual representations are extracted using the pretrained EfficientNetV2B3 model to capture the high-level characteristics of Android applications. Additionally, Locality-Sensitive Hashing (LSH) is employed to extract opcode sequence features that represent low-level, detailed characteristics of the application. These heterogeneous features are then fused using a Multimodal Factorized Bilinear pooling (MFB) algorithm to create a more discriminative representation of the malware. Building on this enhanced feature representation, a Transformer Encoder-based Android Anomaly Detection (TEAAD) model is introduced. By leveraging the transformer architecture, the TEAAD effectively learns to detect anomalies in Android malware. The experimental results demonstrate that the TEAAD model based on fused features outperforms other deep-learning models, achieving a detection accuracy of 96.87%. The MFB feature fusion method exhibits superior malware identification capabilities compared with other research methods.

In recent years, there has been significant progress in terms of accuracy and robustness of deep-learning-based algorithms for object detection that have been widely applied in industry. However, in the field of small object detection, currently used object detection algorithms suffer from high rates of missed detections and false positives. Therefore, in this study, a YOLO small object detection algorithm, viz., BS-YOLO, which is based on SCConv and BSAM attention mechanism, is developed. First, in response to the problem of the large amount of redundant information generated in the feature extraction network, a new module, viz., C3SC, is proposed to reconstruct the backbone network using SCConv. This module reduces redundant information in both spatial and channel aspects of the extracted feature maps, thereby improving the quality of the feature maps extracted by the backbone network, and in turn enhancing detection accuracy. Second, a new attention mechanism, viz., BSAM, is proposed by combining CBAM and the BiFormer self-attention mechanism, by which weights are allocated reasonably in both spatial and channel aspects, making the feature map more focused on effective information and suppressing background interference. Finally, to solve the problem of uneven distribution of difficult and easy samples in terms of small object detection, Slideloss is used to optimize the loss function, thereby improving the effectiveness of the algorithm for small object detection. The experimental results obtained using the RSOD dataset show that the BS-YOLO algorithm has a precision of 94.2%, a recall rate of 91.6%, and a mAP@0.5 of 95.9%, corresponding to improvements of 3.3, 0.1, and 3.6 percentage, respectively, compared to the original YOLOv5 algorithm. This indicates that the BS-YOLO algorithm can effectively improve the accuracy of small object detection and reduce the missed detection rate.

Deep learning has driven the development of artificial intelligence, which is widely used in computer vision. It provides breakthroughs and remarkable results in complex tasks such as image recognition, object detection, object tracking, and face recognition, demonstrating its excellent recognition and prediction capabilities. However, vulnerabilities and loopholes in deep learning models have been gradually exposed. Deep learning techniques, represented by convolutional neural networks, are extremely sensitive to well-designed adversarial examples, which can easily affect the security and privacy of the models. This paper first summarizes the concept of adversarial attacks, reasons for generating adversarial examples, and related terms. It outlines several types of classical adversarial attack strategies in the digital and physical domains and analyzes their advantages and disadvantages. Second, it focuses on computer vision and summarizes the latest research in adversarial attacks during tasks such as object detection, face recognition, object tracking, monocular depth estimation, and optical flow estimation, from both the digital and physical domains, as well as the various datasets commonly used in the study. It also briefly introduces the current stage of adversarial example defense and detection methods, summarizes the advantages and disadvantages of these methods, and describes examples of the applications of adversarial sample defense for various visual tasks. Finally, based on the summary of adversarial attack methods, it explores and analyzes the deficiencies and challenges of existing computer vision adversarial attacks.

The continuous integration of artificial intelligence and robotics technology has facilitated the widespread adoption of multi-rotor Unmanned Aerial Vehicles (UAVs) across various fields, demonstrating their flexibility and efficiency. However, when developing and validating flight control algorithms or solutions for multi-rotor UAVs, researchers face high costs and significant risks. To mitigate these risks and enhance the efficiency of algorithm testing and optimization, simulation platforms for multi-rotor UAVs provide a safe and controlled environment. In this regard, this paper first introduces conventional models of multi-rotor UAVs, selecting the commonly used quadrotor as the representative model. It then elaborates on dynamic models according to different levels of simulation. Subsequently, it provides an overview of the conventional system framework of multi-rotor UAV simulation platforms and discusses their evaluation methods and classification approaches. The evaluation of simulation platforms is detailed from the perspectives of function and performance. multi-rotor UAVs are classified based on whether they support an interactive learning environment and their focus areas: dynamics, sensors, and multi-UAV coordination. This paper also reviews the main solutions for existing UAV flight missions, analyzing typical multi-rotor UAV simulation platforms within the context of traditional and learning-based methods. Finally, the paper outlines future directions for multi-rotor UAV simulation platforms.

Operating Systems (OSs), which form a critical infrastructure in the information age, are widely used in core fields such as medical care, industries, and the military. Their reliability and security directly determine the operational stability of these key fields, while vulnerabilities can lead to serious consequences, including system crashes and data leakage. Hence, the construction of a systematic security assurance system would be of great theoretical and engineering value. This paper systematically reviews the research achievements in this field over the past decade based on the framework of ″formal specification—formal verification—engineering implementation″ and analyzes technical paths and practical applications. With regard to formal specification level, it clarifies the differences between model specifications that describe system functions based on mathematical structures, such as transition systems, and property specifications that define safety and liveness requirements based on Linear Temporal Logic (LTL). This analysis focuses on two key aspects: functional correctness and security attributes. Functional correctness covers task management scheduling, memory allocation and recycling, exception interrupt handling, inter-task communication, and file system read—write consistency, while security attributes focus on the BLP and BIBA models for access control, multidomain isolation in separation kernels, and noninterference and nonleakage theories of information flow. For formal verification, three core methods are considered: deductive proof, which verifies program consistency by relying on Hoare's logic; model checking, which verifies temporal properties based on LTL and Computation Tree Logic (CTL); and standardized process of property verification. Taking the seL4 microkernel, which is the first to achieve functional correctness and information flow non-interference through machine proof, as a case study, this discussion reveals the transformation from theory to engineering. With regard to engineering applications, the achievements in Controller Area Network (CAN) bus communication verification within the automotive field and robustness detection of intercomponent communication in the Android system of smartphones are summarized. This systematic review aims to establish a foundation for future research in related fields, provide dataset support for large language models, and provide a reference for the engineering implementation of these technologies.

This study proposes a Common and Differential Cross-Attention Module-Bird's-Eye View (CDCAM-BEV) algorithm that combines 4D millimeter-wave radar and vision fusion to improve target detection accuracy for pedestrian and vehicle target recognition and localization in autonomous driving scenarios. First, a radar cylinder network is designed to encode the 4D radar point cloud into a pseudo image and convert the monocular image into a Bird's-Eye View (BEV) feature through Orthogonal Feature Transformation (OFT). Second, based on the cross-attention mechanism, a Common Information Extraction Module (CICAM) and a Differential Information Extraction Module (DICAM) are used to fully explore the common and differential information between radar and images. Finally, a BEV feature fusion module is designed based on CICAM and DICAM to achieve feature-level fusion of image and radar information in the BEV space. Experiments are conducted on the VOD dataset, and the CDCAM-BEV algorithm is compared with five other 3D object detection algorithms. The experimental results show that CDCAM-BEV achieves better detection performance in multiple modes. In the 3D mode, the average detection accuracy of CDCAM-BEV is 3.65 percentage points higher than that of the second ranked Part-A²; in the BEV mode, it is 5.04 percentage points higher than that of the second ranked PointPillars; in the Average Directional Similarity (AOS) mode, it is 2.62 percentage points higher than that of the second ranked Part-A². These results show that CDCAM-BEV exhibits excellent performance in all modes, effectively fusing images and 4D radar point cloud features, which significantly improves the accuracy and reliability of object detection.

Large Language Models (LLMs) have made significant progress in dialogue, reasoning, and knowledge retention. However, they still face challenges in terms of factual accuracy, knowledge updates, and a lack of high-quality domain datasets for handling knowledge-intensive tasks in the electricity sector. This study aims to address these challenges by introducing an improved Retrieval-Augmented Generation (RAG) strategy. This strategy combines hybrid retrieval with a fine-tuned generative model for efficient knowledge capturing and updating. The Metadata-driven RAG framework (Meta-RAG) is proposed for knowledge Question Answering (QA) tasks in the electricity domain. This includes data preparation, model fine-tuning, and reasoning retrieval stages. The data-preparation stage involves document conversion, metadata extraction and enhancement, and document parsing. These processes ensure efficient indexing and structured processing of power regulation documents. The Electricity Question Answering (EleQA) dataset, consisting of 19 560 QA pairs, is constructed specifically for this sector. The model fine-tuning stage uses multi-question generation, chain-of-thought prompting, and supervised instruction fine-tuning to optimize the reasoning abilities in specific tasks. The retrieval reasoning stage employs mixed encoding and re-ranking strategies, combining retrieval and generation modules to improve answer accuracy and relevance. Experiments validate the effectiveness of Meta-RAG. Compared to baseline models such as Self-RAG, Corrective-RAG, Adaptive-RAG, and RA-ISF, Meta-RAG shows higher answer accuracy and retrieval hit rates. Meta-RAG with the Qwen1.5-14B-Chat model achieves an overall accuracy of 0.804 3, surpassing the other methods. Ablation and document recall experiments indicate that document retrieval significantly impacts the framework performance, with a 0.292 8 drop in accuracy when the retrieval capability is lost.

Blockchain has gradually evolved into a critical infrastructure that supports the digital economy. However, its inherent characteristics such as anonymity, cross-chain interoperability, and multi-party participation have led to frequent security incidents, including fraud, money laundering, and cyberattacks, which pose serious threats to the stability and compliance of the blockchain ecosystem. Although existing analytical tools and methods have made notable progress in blockchain service security, they suffer from limited generalizability, insufficient reasoning capabilities, and poor adaptability to the evolution of complex business logic. The rapid development of generative Large Language Model (LLM) has significantly reshaped the service computing paradigm. With their strong capabilities in natural language understanding, knowledge reasoning, and multimodal integration, LLM provide new perspectives and technical pathways for research on blockchain service security. This paper systematically reviews the progress of LLM applications in three major areas: pre-event smart contract auditing, in-event anomaly detection, and post-event cross-chain behavior correlation. Further, it summarizes their advantages and limitations and highlights representative practices of LLM-enabled blockchain security. Finally, open research challenges and future directions are discussed, aiming to provide insights for building a trustworthy, interpretable, and efficient framework for blockchain service computing and governance.

Large Language Model (LLM)-based Multi-Agent System (MAS) has demonstrated significant potential in handling complex tasks. Their distributed nature and interaction uncertainty can lead to diverse anomalies that threaten system reliability. This paper presents a comprehensive review, identifying and classifying these anomalies systematically. Seven representative multi-agent systems and their corresponding datasets are selected, accounting for 13 418 operational traces, and a hybrid data analysis method is employed, combining preliminary LLM analysis with expert manual validation. A fine-grained, four-level anomaly classification framework is constructed, encompassing the following anomalies: model understanding and perception, agent interaction, task execution, and external environment. Typical cases are analyzed to reveal the underlying logic and external causes of each type of anomaly. Statistical analysis indicates that model understanding and perception anomalies account for the highest proportion, with ″context hallucination″ and ″task instruction misunderstanding″ being the primary issues. Agent interaction anomalies represent 16.8%, primarily caused by ″information concealment″. Task execution anomalies account for 27.1%, mainly characterized by ″repetitive decision errors″. External environment anomalies account for 18.3%, with ″memory conflicts″ as the predominant factor. In addition, the model perception and understanding of anomalies often act as root causes, triggering anomalies at other levels, highlighting the importance of enhancing fundamental model capabilities. These classification and root cause analyses aim to provide theoretical support and practical reference for building highly reliable LLM-based multi-agent systems.

Post-Training Quantization (PTQ) is an efficient model compression method that converts the parameters of high-precision floating-point models into low-bit integer representations without requiring retraining, using only a small amount of unlabeled calibration data. This method significantly reduces storage and computational overhead while maximizing the retention of the original model's inference accuracy; therefore, it is widely recognized and adopted in both academia and industry. This paper systematically summarizes the progress of research on PTQ from four dimensions: quantization steps, method classification, tool ecosystem, and application advancements. First, a clear framework for the quantization process is constructed, covering steps such as dynamic range statistics, quantization parameter calculation, weight and activation quantization, error optimization, and model generation. Second, a complete classification system for quantization methods is proposed, which includes quantization granularity, bit width, calibration methods, and structure-guided quantization. Third, the tool ecosystem supporting the large-scale application of PTQ is analyzed, and its value in hardware adaptation and engineering deployment is discussed. Finally, this paper summarizes the progress in the integration and application of PTQ methods and highlights practical challenges, particularly those related to cross-modal consistency, extremely low-bit semantic collapse, and hardware adaptation. These challenges not only reveal the limitations of current technologies but also provide important directions for future research. This review provides a reference framework for PTQ methods in academia and industry, thereby facilitating the widespread application of artificial intelligence in resource-constrained scenarios.

Current U-Net-based pavement crack detection methods do not fully consider the interaction between the features of each level of the encoder, causing incomplete detection results or missed detections because of information loss during the downsampling process. To address this issue, this study proposes a pavement crack detection method based on multi-level feature fusion. In the encoding stage, the features of cracks at different levels are extracted to form crack feature representations from shallow to deep layers. In the skip connection section, a cross-level fusion strategy based on an improved Channel Cross Transformer (CCT) is adopted to enhance the complementarity between features at each level and enrich the expression of crack features. In the decoding stage, the feature fusion module is used to optimize the decoder's utilization of encoder features, promote the transmission of crack features, and improve the perception ability of crack features. In a series of comparative and ablation experiments on two public datasets, DeepCrack and CRACK500, the proposed method outperforms six other methods, including DeepCrack and Swin-UNet. On DeepCrack, the proposed method increases the F1 value by 2.30 and 2.51 percentage points, respectively, compared to those of DeepCrack and Swin-UNet, while on CRACK500, it increases by 1.65 and 1.00 percentage points, respectively.

DeepFake-enabled abuse of face forgery technology has given rise to considerable security risks to society and individuals; therefore, DeepFake detection has become a hot topic of research. Current deep learning-based forgery detection techniques exhibit good results on High-Quality (HQ) datasets but show poor performance on Low-Quality (LQ) datasets and across different datasets. To improve the generalization of DeepFake detection performance, this paper proposes a Multi-Scale Dual-Stream Network (MSDSnet) for DeepFake detection. The network input is divided into a spatial-domain feature stream and a high-frequency noise feature flow. First, the Multi-Scale Fusion (MSF) module is used to capture the tampered coarse-grain facial features from images and fine-grained high-frequency noise information from forged images in different situations. The network fully integrates the dual-stream features of the spatial-domain feature stream and high-frequency noise feature flow through the MSF module. The Multi-modal Interaction Attention (MIA) module further interacts to learn the dual-stream information. Finally, a Frequency Channel Attention Network (FcaNet) is used to obtain the global information of the forged face features for complete detection and classification. Experimental results show that the proposed method achieves 98.54% accuracy on the HQ dataset Celeb-DF v2 and 93.11% on the LQ dataset FaceForensics++. Simultaneously, the experimental results are better than those obtained using other methods in cross-dataset experiments.

In recent years, Large Language Models (LLMs) such as GPT, LLaMA, Qwen, and DeepSeek, have achieved significant breakthroughs in natural language processing, computer vision, multimodal learning, and other fields. However, constrained by factors such as their reasoning mechanisms, parameter scales, and the inherent knowledge contained within their training data, these models often suffer from issues like ″hallucinations″—characterized by inaccurate answers and even factual deviations—when handling complex tasks, addressing questions from professional domains, or generating time-sensitive content. These limitations severely hinder their application in high-reliability scenarios. The ″tool learning″ paradigm is attracting increasing attention as a promising solution to these capability bottlenecks. Its primary objective is to enable LLMs to understand and utilize external tools to complete specific tasks. By invoking external tools, such as databases, search engines, and mathematical tools, LLMs can transcend their parameterized knowledge; enhance their reasoning, decision-making, and execution capabilities; and mitigate hallucination problems. This paper systematically reviews the development context and technical advancements in LLM tool learning, analyzes the expansion of LLM capabilities through tools, summarizes tool invocation mechanisms ranging from in-context learning to fine-tuning training, and discusses key issues including performance optimization and adaptive tool generation. The paper also analyzes evaluation methods for LLM tool invocation, summarizes the current challenges in tool learning, and outlines future research directions.

The Siamese tracking network is a popular target tracking framework that includes three modules: backbone, fusion, and positioning networks. The Transformer is a relatively new and effective implementation method for fusion network modules. The encoder and decoder of the Transformer use a self-attention mechanism to enhance the features of the Convolutional Neural Network (CNN). However, the self-attention mechanism can only enhance features in the spatial dimension without considering feature enhancement in the channel dimension. To enable the self-attention network of the Transformer to enhance features both in the spatial and channel dimensions and provide accurate correlation information for the target localization network, a Transformer tracker based on dual-dimensional feature enhancement is proposed to improve the Transformer fusion network. First, using the third- and fourth-stage features of the backbone network as inputs, channel dimension feature enhancement is performed via CAE-Net in the self-attention module of the Transformer encoder and decoder to enhance the importance of the channel. Subsequently, two-stage feature-weighted fusion and linear transformation are performed via SAE-Net to obtain the self-attention factors Q, K, and V. Finally, spatial dimension feature enhancement is performed via a self-attention operation. Experiments conducted on five widely used public benchmark datasets reveal that the improved Transformer feature fusion module can improve the tracking performance of the tracker with minimal reduction in speed of tracking.

When implementing Federated Learning (FL) in a cell-free network environment, user scheduling and resource allocation strategies are crucial for optimizing system time overhead, improving user reachability, and accelerating FL convergence rate. To address the issue of uneven resource allocation, this study designs an optimization scheme that combines user scheduling, CPU processing frequency, and power allocation. This scheme aims to achieve fair resource allocation by maximizing the minimum user rate in the system, thus enhancing FL performance. The joint optimization problem is decomposed into two subproblems: user scheduling and power allocation. For user scheduling, this study proposes a greedy scheduling algorithm based on k-means clustering to comprehensively evaluate channel conditions and data "value" of users and categorize users into different groups. Subsequently, for the resource occupation situation, a personalized CPU processing frequency allocation plan is developed for users within each group based on their resource occupancy. Finally, by independently executing user scheduling within each group, user selection is performed efficiently and precisely, and the complexity of user selection is effectively reduced via early grouping. For power allocation, this study introduces a Bisection Method-based Power Allocation (BM-PA) algorithm. This algorithm not only considers fairness among users but also prioritizes resource-constrained users to ensure that they can obtain superior resource allocation. The BM-PA algorithm achieves fast convergence of power allocation using a low-complexity iterative optimization process, significantly improving the resource utilization efficiency without deteriorating the system performance. In this study, a reasonable user scheduling strategy serves as the foundation for obtaining optimal solutions for the power allocation subproblem. This study adopts an alternating iteration method that allows independent optimization in each subproblem while considering the solution of the other subproblem. Via multiple rounds of iterative optimization, this interdependent relationship ensures that power resources are reasonably allocated to users who need them the most or are most likely to effectively utilize them, thus enhancing the overall system performance. This study realizes joint optimization solutions that significantly improve overall system performance. Simulation results show that compared with the baseline algorithm, the proposed algorithm exhibits outstanding performance in terms of downlink achievable rates-the average improvement reaches up to 103.34% under optimal conditions. Additionally, the uplink achievable rates improve by up to 102.78%. Furthermore, the proposed algorithm can save 67.44% of the FL task training time on average compared to the baseline algorithm, particularly when the FL learning model accuracy reaches 90%, wherein the time overhead of the proposed algorithm is minimal.

Deep learning-based object detection has significantly improved the detection of medium and large targets. However, when detecting small objects, traditional algorithms often face challenges such as missed detections and false positives owing to the inherent issues of small scale and complex backgrounds. Therefore, this study aims to enhance the accuracy of small object detection by improving the YOLOv8 model. First, the convolutional module in the backbone is replaced with the RFAConv module, which enhances the ability of the model to process complex images. Second, a Mixed Local Channel Attention (MLCA) mechanism is introduced in the neck part, allowing the model to fuse features from different layers more efficiently while maintaining computational efficiency. Third, the Detect head of YOLOv8 is replaced with the Detect_FASFF head to address the inconsistency between different feature scales and improve the ability of the model to detect small objects. Finally, the Complete Intersection over Union (CIoU) loss function is replaced with the Focaler-IoU loss function, enabling the model to focus more on small objects that are difficult to locate precisely. Experimental results show that the improved model increases mAP@0.5 by 4.8 percentage points and mAP@0.5:0.95 by 3.0 percentage points on the FloW-Img dataset, which is sparse in small objects. On the VisDrone2019 dataset which has a high density of small objects, mAP@0.5 increases by 5.9 percentage points and mAP@0.5:0.95 improves by 4.0 percentage points. In addition, generalization comparison experiments are conducted on the low-altitude dataset AU-AIR and the pedestrian-dense detection dataset WiderPerson. The optimized model significantly improves the accuracy of small object detection compared with the original model and expands its applicability.

As a mainstream tool for container orchestration, Kubernetes can support automatic deployment, service discovery, and load balancing. It is known for its high availability and performance. However, scheduling strategies such as the best adaptation algorithm or the minimum negative cut method ignore the heterogeneity and energy differences of nodes. In addition, Kubernetes tools only consider CPU and memory resources and set a unified weight mechanism in advance, which can easily lead to problems such as load imbalance, performance degradation, and the inability to satisfy refined scheduling. To address these existing problems, this study proposes a heterogeneous task scheduling algorithm based on multi-dimensional resources, namely A-KCSS. The A-KCSS algorithm is based on the heterogeneous computing resources of a cluster. It adds disk Input/Output (I/O), network I/O load, and GPU resources as evaluation indicators for filtering and screening, and it more comprehensively considers the heterogeneity of nodes. This study also introduces a weight calculation model based on multi-dimensional resource factors. Based on the resource requirements of the task to be scheduled, the weight value of each dimension of the resource factor of the task to be scheduled is calculated, and the score of each node is calculated based on the real-time resource utilization of the cluster. Nodes are prioritized according to the score, and the node with the highest priority is selected for scheduling. The performance of the A-KCSS algorithm is experimentally verified on the Kubernetes cluster. Compared with the default scheduling algorithm and the KCSS algorithm, the average response time is reduced by 10% and 4%, the throughput is increased by 30% and 15%, the availability is improved by 40% and 30%, and the load balancing performance is increased by 23% and 18%, respectively, thereby improving the overall cluster performance.

In the context of ongoing advancements in educational informatization, constructing precise and efficient curriculum knowledge graphs has become key to promoting personalized education development. As a structured knowledge representation model, curriculum knowledge graphs reveal complex relations between curriculum content and learning objectives to optimize the allocation of educational resources, and tailoring personalized learning paths for learners. This survey presents a discussion around the techniques used to construct curriculum knowledge graphs, starting with an explanation of the basic concepts; intrinsic connections; and significant differences among general, educational, and curriculum knowledge graphs. It then delves into the key technologies used for building curriculum knowledge graphs, covering aspects such as curriculum ontology design, entity extraction, and relation extraction, and provides a detailed analysis and summary of their evolution, key features, and limitations. Furthermore, it explores the application value of curriculum knowledge graphs in scenarios such as learning resource recommendation, learner behavior profile and modeling, and multimodal curriculum knowledge graph construction. Finally, it focuses on the challenges in constructing curriculum knowledge graphs, such as data diversity and heterogeneity, difficulties in quality evaluation, and the lack of cross-curriculum integration, and provides future-oriented insights based on cutting-edge technologies such as deep learning and Large Language Models (LLMs).

As multivariate time series data become increasingly prevalent across various industries, anomaly detection methods that can ensure the stable operation and security of systems have become crucial. Owing to the inherent complexity and dynamic nature of multivariate time series data, higher demands are placed on anomaly detection algorithms. To address the inefficiencies of existing anomaly detection methods in processing high-dimensional data with complex variable relations, this study proposes an anomaly detection algorithm for multivariate time series data, based on Graph Neural Networks (GNNs) and a diffusion model, named GRD. By leveraging node embedding and graph structure learning, GRD algorithm proficiently captures the relations between variables and refines features through a Gated Recurrent Unit (GRU) and a Denoising Diffusion Probabilistic Model (DDPM), thereby facilitating precise anomaly detection. Traditional assessment methods often employ a Point-Adjustment (PA) protocol that involves pre-scoring, substantially overestimating an algorithm's capability. To reflect model performance realistically, this work adopts a new evaluation protocol along with new metrics. The GRD algorithm demonstrates F1@k scores of 0.741 4, 0.801 7, and 0.767 1 on three public datasets. These results indicate that GRD algorithm consistently outperforms existing methods, with notable advantages in the processing of high-dimensional data, thereby underscoring its practicality and robustness in real-world anomaly detection applications.

Password leakage incidents often involve the leakage of user passwords and identity information. Because users are accustomed to reusing passwords across multiple network services, attackers can tweak leaked passwords to accurately attack user accounts. This is called a credential tweaking attack. By analyzing large-scale leaked passwords and the corresponding user identity information, this study finds that user strategies for creating passwords are often associated with user identity information. However, current research on credential tweaking attacks relies only on leaked password structures and ignores leaked user identity information when predicting password tweaking strategies. To improve the accuracy of credential tweaking attacks, this study designs a credential tweaking attack optimization method based on user identity information. In the preprocessing phase, username and regional information is extracted from the user identity information and the probability of users' different password creation strategies in different regions is statistically calculated. In the training phase, regional information is combined to learn users' character-level editing operations on leaked passwords. In the password generation phase, a password generation method that integrates character-level editing operations, structure-level editing operations, and username information is designed. The experimental results show that in an attack with 10³ guesses, the cracking rate of this method has a maximum improvement of 41.8% compared to the existing best method (PassBERT), highlighting the threat posed by credential tweaking attacks based on user identity information to password security.

This study presents a facial emotion recognition network based on UniRepLKNet to address the difficulty in effectively capturing feature information and preventing key facial information from occupying a more prominent position in the facial emotion recognition process. Moreover, to extract facial emotional features more accurately, the study designs a masked polarized self-attention module that combines U-Net and a polarized self-attention mechanism. This module can deeply mine the dependency between channels and spaces. It can also strengthen the influence of local key information of the face on emotion recognition through a multi-scale feature fusion strategy. The study optimizes UniRepLKNet, a universal large kernel Convolutional Neural Network (CNN), and proposes the EmoRepLKNet neural network structure. In EmoRepLKNet, the mask-polarized self-attention module enables the network to extract key information for facial emotion recognition. Combined with the wide receptive field of large kernel CNN, facial emotions can be recognized effectively. Experimental results show that on the facial emotion recognition dataset FER2013, EmoRepLKNet achieves an accuracy of 76.20%, outperforming existing comparison models and significantly improving facial emotion recognition accuracy compared to that of UniRepLKNet. Additionally, on the single-label portion of the RAF-DB dataset, the proposed method achieves an accuracy of 89.67%.

Adversarial examples can be used to perform transferable attacks on black-box models using surrogates, without knowing the internal structure and parameters of the black-box model. Previous studies have reported relatively low transferability of targeted attacks on black-box models. This study proposes a method for enhancing the transferability of image-directed targeted attacks based on feature fusion. First, adversarial examples are generated via ensemble attacks. Subsequently, using the gradient direction of existing adversarial examples as a baseline, clean features extracted from the original image are used as perturbations to fine tune the existing adversarial examples for improving the transferability of targeted attacks. For model ensembling, a gradient adaptive module is introduced based on the contribution of each model to the overall adversarial objective. To reduce the gradient differences among different models, a gradient filter is proposed for synchronously controlling the gradient direction. Using the feature fusion module, the clean features of the original image are mixed to fine tune the gradient direction of the existing adversarial examples for mitigating the issue of overfocusing on specific features. Experiments on the ImageNet-Compatible dataset reveal that, compared to the Clean Feature Mixup (CFM) method, the proposed method improves the average attack success rate by 7.7 percentage points for non-robustly trained models and by 5.3 percentage points for robustly trained and Transformer models, demonstrating the effectiveness of the method.

Traditional machine learning algorithms perform well only when the training and testing sets are identically distributed. They cannot perform incremental learning for new categories or tasks that were not present in the original training set. Continual learning enables models to learn new knowledge adaptively while preventing the forgetting of old tasks. However, they still face challenges related to computation, storage overhead, and performance stability. Recent advances in pre-training models have provided new research directions for continual learning, which are promising for further performance improvements. This survey summarizes existing pre-training-based continual learning methods. According to the anti-forgetting mechanism, they are categorized into five types: methods based on prompt pools, methods with slow parameter updating, methods based on backbone branch extension, methods based on parameter regularization, and methods based on classifier design. Additionally, these methods are classified according to the number of phases, fine-tuning approaches, and use of language modalities. Subsequently, the overall challenges of continual learning methods are analyzed, and the applicable scenarios and limitations of various continual learning methods are summarized. The main characteristics and advantages of each method are also outlined. Comprehensive experiments are conducted on multiple benchmarks, followed by in-depth discussions on the performance gaps among the different methods. Finally, the survey discusses research trends in pre-training-based continual learning methods.

Advances in computational power and network technologies have driven robots toward miniaturization, swarm intelligence, and autonomous capabilities. Robot software deployed on robotic hardware must integrate diverse modules from low-level device drivers and controls to high-level motion planning and reasoning, resulting in increasingly complex architectures. A communication and programming framework for multi-robot systems—focusing on standardization, modularization, and platformization—can alleviate the complexity of programming robotic software. The development trends in robotic software and hardware architecture show that a swarm robotic system is a multi-domain, heterogeneous, and distributed system composed of computing nodes, actuators, sensors, and other hardware devices interconnected through wired or wireless networks. The heterogeneity of hardware devices makes it difficult to integrate software components into a single framework. This survey summarizes and analyzes existing robotic communication frameworks in terms of ease of use and portability, comparing their core features, such as programming models, heterogeneous hardware support, communication and coordination mechanisms between components, and programming languages. The survey then highlights the technical trends of advanced topics such as real-time virtualization, component orchestration, and fault tolerance. Moreover, this survey focuses on building a next-generation framework on a meta Operating System (OS) foundation, aiming to build a ubiquitous and integrated multi-robot software architecture for human-machine-object interactions.

This study explores the use of Artificial Intelligence (AI) technology throughout the neutron scattering experiments′ lifecycle to determine how AI technology can revolutionize key aspects such as experimental apparatus, data acquisition, and data processing. The study begins by introducing the fundamental principles and experimental procedures of neutron scattering technology before focusing on the multifaceted applications of AI technology in neutron scattering experiments. These applications include optimizing experimental infrastructure, data acquisition, and imaging preprocessing, as well as characterizing experimental samples in neutron diffraction, neutron reflection, and Inelastic Neutron Scattering (INS). This study demonstrates the importance of AI technology in increasing the intelligence level of experiments, accelerating data processing, and improving the accuracy and reliability of data analyses. In addition, an in-depth discussion is held on the future application of AI technology in neutron scattering experiments, indicating that with the continuous advancement of technologies such as multimodal learning, interpretable models, large language models, and AI-Ready databases, AI technology is poised to bring revolutionary changes to neutron scattering experiments, opening up new avenues for revealing the microstructure and properties of complex material systems.

Accurate photovoltaic power prediction is crucial for enhancing grid stability and improving energy utilization efficiency. To address the limitations of existing methods, which struggle to simultaneously consider both long-term dependencies and short-term variation patterns of photovoltaic power, this study proposes a novel photovoltaic power prediction method named Solarformer. This method integrates a Pyramid Attention Module (PAM) with a Temporal Convolutional Network (TCN) to optimize the Transformer architecture. First, multiple feature selection mechanisms are employed to screen the input features, to enhance the model′s ability to characterize photovoltaic data features. Second, a coarse-grained construction module and PAM are utilized to optimize the Transformer encoder, capturing the long-term temporal dependency features of photovoltaic power at multiple scales. Third, a constraint mechanism based on the sunrise-sunset effect of photovoltaic power and the TCN are employed to optimize the Transformer decoder, strengthening the model′s ability to capture short-term variation features of photovoltaic power and better model its short-term variation patterns. Experimental results on the Sanyo dataset from Australia demonstrate that Solarformer can effectively improve photovoltaic power forecasting accuracy. Compared with the DLinear model, it reduces the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Symmetric Mean Absolute Percentage Error (SMAPE) by approximately 7.45%, 6.99%, and 14.10%, respectively.