Semi-supervised multi-label learning (SSMLL) aims to address the challenge of limited labeled data in multi-label learning (MLL) by leveraging unlabeled data to improve the model’s performance. While pseudo-labeling has become a dominant strategy in SSMLL, most existing methods assign equal weights to all pseudo-labels regardless of their quality, which can amplify the impact of noisy or uncertain predictions and degrade the overall performance. In this paper, we theoretically verify that the optimal weight for a pseudo-label should reflect its correctness likelihood. Empirically, we observe that on the same dataset, the correctness likelihood distribution of unlabeled data remains stable, even as the number of labeled training samples varies. Building on this insight, we propose Distribution-Calibrated Pseudo-labeling (DiCaP), a correctness-aware framework that estimates posterior precision to calibrate pseudo-label weights. We further introduce a dual-thresholding mechanism to separate confident and ambiguous regions: confident samples are pseudo-labeled and weighted accordingly, while ambiguous ones are explored by unsupervised contrastive learning. Experiments conducted on multiple benchmark datasets verify that our method achieves consistent improvements, surpassing state-of-the-art methods by up to 4.27%.

Partial multi-label learning (PML) aims to train a classifier on dataset whose instances are over-annotated with not only relevant labels but also irrelevant labels, which is common when datasets are collected from crowd-sourcing platform. Existing works primarily approach it from a curriculum learning perspective, leveraging the memorization effect to disambiguate noisy labels and produce robust predictions. However, these methods are based on non-adaptive weighting functions and lack theoretical guidance for optimal weighting. To overcome these issues, a calibrated disambiguation model named PML-CD is proposed. We firstly formulate the optimal weighting function for curriculum-based disambiguation, which is equivalent to the calibration of the model's predicted confidences, thus provide a guidance for curriculum designing. To obtain the optimal weighting function from PML dataset during the training, a transferable calibrator is designed, which takes the histogram of positive samples' confidences as input, and outputs the optimal curriculum weighting for training. Prototype alignment regularization is also proposed to promote the model's performance. Experiments conducted on Pascal VOC, MS-COCO, NUS-WIDE and CUB have verified that our method outperforms existing state-of-the-art PML methods.

In this paper, we propose a UAV swarm clustering routing optimization algorithm based on a dual-network architecture. To address the issue of unstable ad-hoc networks caused by the large number of UAV swarms and rapid speed changes, we first establish a hierarchical UAV swarm model including cluster heads, cluster members, and isolated nodes, and propose a communication link quality evaluation mechanism that comprehensively considers multiple communication factors. For the performance optimization problem of drones, we establish decentralized partial observable Markov decision process (Dec-POMDP). For UAV nodes, we design a multi-agent deep Q-learning algorithm for cluster head election. In the Swarm-DQN algorithm, a dual-network structure of an action network and a target network is introduced to solve the over-estimation problem of multiple UAVs. In terms of cluster management, we propose a method based on velocity and position similarity, enabling cluster members to join the most suitable cluster head. Experimental results show that the proposed Swarm-DQN algorithm outperforms traditional methods in terms of packet delivery rate, throughput, and average remaining energy, thereby validating its effectiveness.

Orthogonal time frequency space (OTFS) modulation has shown exceptional performance in high-mobility scenarios. However, downlink channel estimation in massive multiple-input multiple-output (MIMO) OTFS systems remains challenging due to limited Doppler and angular resolutions, underutilization of domain-specific channel characteristics, and substantial computational complexity. In this paper, we propose a multi-domain decoupled channel estimation framework that separates the coupled channel into its individual domain components. The framework enhances the utilization of channel characteristics and provides more flexibility in algorithm design. Based on the proposed framework, we develop a multi-domain decoupled Bayesian learning (MDDC-BL) channel estimation algorithm, achieving super-resolution performance in both Doppler and angular domains. For angular-domain estimation, we devise a dictionary learning based pattern-coupled Bayesian learning (DL-PC-BL) channel estimation algorithm to effectively capture the channel characteristics and handle the off-grid components. Additionally, the MDDC-BL algorithm reduces the dimensionality of the estimation model, thereby significantly lowering computational complexity. Simulation results demonstrate the superior performance of the proposed framework and channel estimation algorithms over existing state-of-the-art schemes.

In this paper, the weighted fractional Fourier transform (WFRFT) signal domain is introduced to enhance the security performance of two-way trusted relay systems at the signal level. The proposed scheme, which requires only a single relay node, leverages the multi-component energy distribution characteristics of WFRFT signals to improve security with low complexity and high power efficiency. The inherent security mechanism of WFRFT analyzed in this paper can be simply summarized as follows: the superposition of components in WFRFT signals that do not satisfy specific constraints will result in the inability to perfectly reconstruct the message signal. Based on this, confidential information is encoded into WFRFT signals with private transform orders, allowing legitimate users to achieve perfect decoding. Since WFRFT signals exhibit energy concentration only in specific transformation domains, mismatched transform orders adopted by the eavesdropper cause energy loss in the information-bearing signal, leading to inter-component interference that further degrades the quality of the recovered signal. The advantages of the proposed scheme in limiting information leakage and improving the achievable secrecy sum rate (SSR) are analyzed. Numerical results validate the theoretical analysis and demonstrate the secrecy performance of the proposed scheme.

In this paper, we investigate hybrid precoder design to enhance the physical layer security of the millimeter-wave (mmWave) system with multiple eavesdroppers. Differing from traditional spatial domain-based schemes, we introduce an extra signal domain dimension and propose a multi-component-based security mechanism for the hybrid precoding system model. The model integrates both signal and spatial domains for designing baseband and analog precoders, ensuring that the received signal at the legitimate user complies specific signal domain characteristics. Consequently, all received energy becomes available for signal reconstruction, whereas the distorted signal at eavesdroppers inevitably forms inter-component interference. By introducing general multi-fractional Fourier transform (GMFRFT) signal domain into the model, we propose a specific GMFRFT-based scheme for subarray architectures. The secrecy performance is analyzed, and power allocation methods between multiple components are designed. Additionally, to optimize the utilization of spatial domain resources, we propose a GMFRFT-based scheme with dynamic subarrays, along with a low-complexity subarray partitioning algorithm. Furthermore, a GMFRFT-based scheme for fully connected architectures is investigated. It is verified to be equivalent to a lower-complexity weighted fractional Fourier transform (WFRFT)-based analog beamforming scheme, the efficacy of which hinges on the adherent secrecy provided by WFRFT signals. Both theoretical analysis and numerical results demonstrate the effectiveness of the proposed schemes.

In this paper, a fast beam training method based on geometric beam patterns is proposed. It accumulates precise geometric patterns created by extremely large-scale MIMO with displacement from different training frames and covers all directions. By analyzing beam gains of different patterns which contributed by the channel angle and geometric shapes, the proposed method acquires more channel angle information and significantly reducing beam training frames. Both analytical and simulation results demonstrate the proposed method has a faster error convergence speed compared with traditional methods. It becomes possible to achieve fast beam alignment in a few training frames for millimeter-wave extremely large-scale MIMO in high-mobility scenarios such as vehicular networks.

Network embedding has been widely investigated to learn low dimensional nodes representation of networks, and serves for many downstream machine learning tasks. Previous network embedding studies mainly focus on static networks, and cannot adapt well to the characteristics of dynamic networks which are evolving over time. Some works on dynamic network embedding have tried to improve the computation efficiency for incremental updates of embedding vectors, while others have made efforts to utilize temporal information to enhance the quality of embedding vectors. However, few existing works can fulfill both efficiency and quality requirements. In this article, a novel dynamic network embedding model named TPANE (Temporal Path Adjacency Matrix based Network Embedding) is proposed. It employs a new network proximity measure: Temporal Path Adjacency, which is capable of capturing the temporal dependency between edges as well as being incrementally computed in an efficient way. It evaluates the similarity between nodes via the count of temporal paths between them, rather than making random sampling approximation, and adopts matrix factorization to obtain embedding vectors. Link prediction experiments on various real-world dynamic networks have been conducted to show the superior performance of TPANE against other state-of-the-art methods. Time consumption analysis also shows that TPANE is more efficient in incremental updates.

The end goal of source localization problem is to find the parameters of the brain source. The source number is also the case of these parameters. In this paper we consider two system identification methods (e.g., the information criterion method and the F-test method) to determine the dipole number only with one EEG or MEG topography. Then compare the results from these two methods and find the better one. The following investigations are presented to show that the information criterion method used in this paper is an advanced approach for determining the dipole number just with one EEG or MEG topography.