Despite neural networks’ progress in musical instrument recognition, the collaborative dynamics in musical ensembles are underexplored. Research on instrumental timbre persistence and inter-instrument coordination in polyphonic soundscapes lacks attention. Genre-specific instrumental combinations have predictable label correlation patterns that can improve recognition if exploited. This paper proposes a Correlation Graph Attention Network with supervision augmentation (CGAT) for instrument recognition, unifying temporal continuity modeling and label correlation-based semantic reasoning. CGAT first generates time-frequency representations of audio signals, then constructs correlation graphs: (1) dynamic temporal correlation graphs for temporal evolution patterns, (2) label correlation graphs for inter-instrument correlation. It integrates music genre information to enhance label semantics. Using instrument and genre labels and the dual graphs, the network creates a graph-attention fusion recognition pipeline via a deep graph attention network (GAT) to optimize feature alignment and accuracy. Experiments on the OpenMIC and a customized dataset validate CGAT’s efficacy.

Partial Multi-label Learning (PML) is a typical weakly supervised learning paradigm. In the complex label noise environment, PML models need semantic correlation features to build an adaptive perception for heterogeneous noise levels of labels. Nevertheless, most existing mainstream PML methods adopt a fixed-threshold strategy for label information propagation, and this strategy exacerbates the bottleneck of label ambiguity, making it unable to adapt to differentiated sample noise scenarios. To address the above situation, this paper proposes a novel label correlation-driven partial multi-label learning algorithm with dynamic noise threshold learning (PML-LE). The PML-LE realizes the embedding of label semantic features and correlation features through the label correlation modeling module to provide semantic support for the subsequent label disambiguation process. Meanwhile, PML-LE designs an adaptive noise thresholding mechanism based on the information entropy of candidate label sets, which dynamically adjusts the pruning strictness of label adjacency matrices according to each sample’s noise level. Experimental results on benchmark datasets demonstrate that the proposed PML-LE algorithm achieves state-of-the-art performance for PML.

Deep clustering refers to the integration of deep learning techniques with clustering. With the popularity of social media, deep clustering for image data has attracted widespread attention from researchers. This paper provides a comprehensive review of the latest research image clustering based on deep learning. First, image clustering is categorized into two main types: self-supervised learning-based and semi-supervised learning-based methods. The paper analyzes and summarizes the methods in each category, highlighting their advantages and disadvantages. Next, the paper compares and analyzes representative methods through experimental results, providing insights for future research and practical applications. Finally, the paper discusses existing challenges and presents potential directions for future research.

Image clustering is a crucial but open and challenging task in machine learning and computer vision. Deep image clustering methods have made significant advancements in largescale and high-dimensional image datasets, but they only explore clusters according to feature similarity. How to improve the semantic plausibility of these clusters remains a challenging problem. To address this problem, we propose a joint semantic image clustering (SIC) with deep learning framework. Our key idea is to explore semantic clusters from both instance-level and clusterlevel perspectives. At the cluster level, we assume that the highest-confidence partitions possess reasonable semantic divisions; at the instance level, we assume that neighboring samples belong to the same semantic category. With the constraints of the cluster level and instance level, the intracluster compactness and interclass discrepancy are increased. Clustering results heavily rely on the quality of image representations. Therefore, we introduce strong data augmentations with three shared-weight backbone networks to learn the most inherent features. The experimental results on various image datasets demonstrate our framework’s superiority over a wide range of state-of-the-art approaches.

Linear autoencoder models with nonnegative constraints and L1 regularization, such as the sparse linear method (SLIM), have shown remarkable performance while maintaining interpretability. However, their practicality is limited by computationally expensive training processes. This paper proposes a simple yet effective training framework for nonnegative and sparse linear autoencoders. We first develop a simple iterative update scheme (IUS) for SLIM and present a theoretical analysis of its convergence and correctness. To enhance computational efficiency, we then introduce a filtering step that prunes insignificant parameters at each iteration in practice. Based on this training scheme, we derive two model variants by removing the zero-diagonal constraint and utilizing random dropout denoising to replace L2 regularization (i.e., the dropout-based regularization in DLAE), respectively. Experimental results demonstrate that the proposed IUS algorithm reduces training time by 53.8-68.5% and memory usage by 55.6% compared to the alternating direction method of multipliers (ADMM) across six benchmark datasets. The proposed model variants achieve comparable or superior performance to state-of-the-art collaborative filtering models on all real-world datasets. These findings validate the proposed training framework's capability to enable feasible deployment of SLIM-like models in efficiency-critical and resource-constrained environments.

Deep attribute graph clustering aims to reveal the underlying structure of attribute graph data and subsequently partition nodes into different clusters. In recent years, thanks to the powerful representation learning capabilities of graph neural networks, significant progress has been made in attribute graph clustering research. However, existing graph neural network-based clustering methods often overly rely on topological structures, thereby compromising the similarity of nodes in the original feature space and leading to suboptimal clustering results. In response to this issue, this paper proposes a feature fusion-based attribute graph clustering method. Its core idea is to fully exploit the information inherent in both topological structures and attribute features, utilizing an attention mechanism to adaptively learn fusion weights. To mitigate the interference of noise neighbors in the information aggregation process, a self-learning module is introduced to enhance trust in information originating from the nodes themselves. Finally, by incorporating a feature constraint module, it contributes to learning more enriched node representations. Experimental results demonstrate that, on four commonly used graph datasets, namely ACM, DBLP, CITE, and AMAP, the proposed method outperforms seven benchmark methods, including GAE, DAEGC, MVGRL, SDCN, AGCN, and DFCN, showing a further improvement in clustering performance and exhibiting a certain superiority.

Recently, communication learning has emerged as a successful tool for tackling complex tasks in multi-agent reinforcement learning (MARL). Many MARL methods use graph neural networks (GNNs) to build a communication learning framework where agents and communication channels can be represented as nodes and edges in a graph. However, most GNN-based MARL methods simply aggregate features of neighboring agents to obtain message representations, which may not extract enough useful information. To tackle this problem, this paper investigates how to extract expressive information from neighboring agents to obtain high-quality message representations. Inspired by the recent success of contrastive learning methods, in this paper, we propose a multi-agent communication protocol via graph contrastive learning (MAGE), which utilizes contrast objectives to learn optimal message representations, considering feature and topological level. At the feature level, we corrupt agent features by adding more noise to insignificant neighboring agent features, to encourage the agent to recognize significant information. At the topological level, we adaptively remove edges by assigning larger removal probabilities to insignificant edges to highlight significant communication structures. Experiments across diverse benchmarks confirm that MAGE outperforms existing GNN-based MARL methods.

As a new paradigm that integrates clustering with federated learning, federated clustering (FC) has recently attracted increasing attention, as it addresses the practical issue of privacy protection in distributed data. In this paper, we provide a comprehensive survey of recent advances in FC. This survey is organized into four parts. First, since FC is often developed by extending existing clustering methods, we review several classical clustering paradigms. Meanwhile, the inherent challenges of FC are summarized, and common improvement strategies are categorized. Second, we summarize experimental setups and evaluation protocols used in FC studies. Third, from the perspectives of data partitioning schemes and whether deep representation learning is incorporated, FC methods are divided into four categories, and representative algorithms in each category are reviewed. Finally, we discuss the limitations of current FC approaches and highlight potential directions for future research.

The density peak clustering algorithm is non-iterative and capable of clustering arbitrarily shaped clusters. However, it also has some shortcomings: it is difficult to correctly select density peaks from decision graphs on complex datasets, resulting in poor clustering performance; The sample allocation process is prone to chain errors; And there is a lack of unified selection criteria for truncation distance parameters. To address these issues, this paper proposes a Density Peak Clustering algorithm based on Density Conflict Domains (DPC-DCD). This method integrates the improved concept of mutual neighbors and introduces the concept of density conflict domain. By removing low-density nodes from the conflict domain, a sufficient number of density peak regions are obtained, replacing the traditional DPC method of calculating density based on truncation distance and manually selecting peaks based on a two-dimensional decision graph. In addition, the algorithm adopts a label propagation mechanism to allocate boundary points. Comparative experiments were conducted with seven improved DPC methods on 16 synthetic datasets and 12 UCI real datasets. The results showed that DPC-DCD can clearly characterize density peak regions and achieve excellent clustering performance on various complex data structures without introducing additional hyperparameters, achieving a fully automated clustering process. The method proposed in this article not only overcomes the dependence of traditional DPC on manual intervention and parameter setting, but also provides a robust and efficient clustering approach for processing data with manifolds, multiple peaks, and uneven density distributions.

Emotion recognition in conversations (ERC) has gained increasing interest in the field of human-computer interactions. While most existing studies still primarily focus on modeling the intrinsic emotional dynamics within dialogues, this approach exhibits inherent limitations in interpreting implicit emotional cues that often rely on commonsense knowledge. Furthermore, a naive integration of external knowledge risks introducing noise and distorting the original semantic representations of the dialogue. Therefore, to effectively utilize external knowledge to enhance emotional understanding, we propose a Knowledge-Enhanced Heterogeneous Graph for Emotion Recognition in Conversation (EmoKEG). Specifically, EmoKEG models contextual semantics and external knowledge as heterogeneous nodes, incorporating a bidirectional knowledge aggregation mechanism to enable dynamic knowledge filtering and interaction. The framework further employs Graph-Transformer for structured reasoning over the constructed graph. To mitigate noise from external knowledge, a gated knowledge-context fusion mechanism is introduced, ensuring effective integration of emotional information. Extensive experimental results on the IEMOCAP and MELD datasets demonstrate the effectiveness of the proposed model.

Multimodal Emotion Recognition in Conversation (MERC) aims to recognize emotions in conversational videos by fusing multimodal information. Current methods mainly rely on the direct fusion of emotional representations. This simple concatenation or linear combination not only fails to fully explore the deep semantic correlations and cross-modal dependencies among visual, textual, audio and other modalities, but also causes redundant feature interference due to the heterogeneity and misalignment of features from different modalities. Therefore, this paper proposes the Graph Dynamic Fusion Network with Modality Mutual Information Maximization (MMGDFN), designed to effectively exploit the complementarity among modalities in conversations. Specifically, by constructing dynamic graph networks to model the interaction relationships among features of various modalities, the method adaptively captures cross-modal dependencies through graph structures. Meanwhile, a mutual information maximization mechanism is introduced to quantify and enhance complementary information among modalities, enabling adaptive balancing of multimodal complementary information and effective filtering of redundant features. Experimental results on two publicly available datasets (i.e. IEMOCAP and MELD) demonstrate that the proposed MMGDFN significantly outperforms state-of-the-art multimodal emotion recognition methods.

Employing graph neural networks (GNNs) for graph clustering has shown promising results in deep graph clustering (DGC). However, existing methods disregard the reciprocal relationship between representation learning and structure augmentation: the more homogeneous the graph, the more cohesive the node representations; the more cohesive the node representations, the more reliable the structure augmentation becomes. Moreover, the generalization ability of existing GNN-based models on the low homophily graph is relatively poor. To this end, we propose a graph clustering framework named synergistic deep graph clustering network (SynC). SynC employs a transform input graph autoencoder (TIGAE) to obtain high-quality embeddings via mitigating the representation collapse issue of GAE for guiding structure augmentation. Then, we recapture neighborhood representations on the refined graph to obtain clustering-friendly embeddings and conduct self-supervised clustering. Notably, these two stages share weights, resulting in synergistic boosting while significantly reducing the number of model parameters. Additionally, we introduce a structure fine-tuning (SF) strategy to improve the model's generalization on the low homophily graph. Extensive experiments on benchmark datasets demonstrate the superiority of SynC. The code is released at https://github.com/Marigoldwu/SynC.

Emotion Recognition in Conversation (ERC) aims to detect the emotions of individual utterances within a conversation. Generating efficient and modality-specific representations for each utterance remains a significant challenge. Previous studies have proposed various models to integrate features extracted using different modality-specific encoders. However, they neglect the varying contributions of modalities to this task and introduce high complexity by aligning modalities at the frame level. To address these challenges, we propose the Multi-modal Anchor Gated Transformer with Knowledge Distillation (MAGTKD) for the ERC task. Specifically, prompt learning is employed to enhance textual modality representations, while knowledge distillation is utilized to strengthen representations of weaker modalities. Furthermore, we introduce a multi-modal anchor gated transformer to effectively integrate utterance-level representations across modalities. Extensive experiments on the IEMOCAP and MELD datasets demonstrate the effectiveness of knowledge distillation in enhancing modality representations and achieve state-of-the-art performance in emotion recognition. Our code is available at: https://github.com/JieLi-dd/MAGTKD.

Due to variations in imaging conditions, images often exhibit discrepancies in color reproduction. Furthermore, motion-induced blur can lead to edge degradation, making color sensitivity and edge blurriness two prevalent and challenging issues in both natural image processing and medical image analysis. To address these challenges, we propose a model termed the Three-View Consistency Momentum Contrastive with Sobel Operator (SVCMC). Specifically, we first design a three-view momentum-update architecture that employs a Sobel-augmented ResNet as the backbone. We then introduce a novel contrastive loss, referred to as the Three-View Consistency Momentum Contrastive Loss. Next, to mitigate the oscillations and slow convergence commonly observed in contrastive learning, we construct a dynamic contrastive loss function that adapts in real time over the training process. Finally, we validated the superiority of our model on two medical image datasets and one natural image dataset, where its classification accuracy and convergence speed significantly outperformed existing state-of-the-art contrastive models.

Communication learning is an important research direction in the multiagent reinforcement learning (MARL) domain. Graph neural networks (GNNs) can aggregate the information of neighbor nodes for representation learning. In recent years, several MARL methods leverage GNN to model information interactions between agents to coordinate actions and complete cooperative tasks. However, simply aggregating the information of neighboring agents through GNNs may not extract enough useful information, and the topological relationship information is ignored. To tackle this difficulty, we investigate how to efficiently extract and utilize the rich information of neighbor agents as much as possible in the graph structure, so as to obtain high-quality expressive feature representation to complete the cooperation task. To this end, we present a novel GNN-based MARL method with graphical mutual information (MI) maximization to maximize the correlation between input feature information of neighbor agents and output high-level hidden feature representations. The proposed method extends the traditional idea of MI optimization from graph domain to multiagent system, in which the MI is measured from two aspects: agent features information and agent topological relationships. The proposed method is agnostic to specific MARL methods and can be flexibly integrated with various value function decomposition methods. Considerable experiments on various benchmarks demonstrate that the performance of our proposed method is superior to the existing MARL methods.

Density peaks clustering (DPC) is a popular clustering algorithm, which has been studied and favored by many scholars because of its simplicity, fewer parameters, and no iteration. However, in previous improvements of DPC, the issue of privacy data leakage was not considered, and the "Domino" effect caused by the misallocation of noncenters has not been effectively addressed. In view of the above shortcomings, a horizontal federated DPC (HFDPC) is proposed. First, HFDPC introduces the idea of horizontal federated learning and proposes a protection mechanism for client parameter transmission. Second, DPC is improved by using similar density chain (SDC) to alleviate the "Domino" effect caused by multiple local peaks in the flow pattern dataset. Finally, a novel data dimension reduction and image encryption are used to improve the effectiveness of data partitioning. The experimental results show that compared with DPC and some of its improvements, HFDPC has a certain degree of improvement in accuracy and speed.

Multi-view self-supervised learning leverages data from multiple independent perspectives to learn features through self-supervised methods. However, this approach faces challenges, such as effectively integrating features from different views and maximizing the selection of negative samples. Meanwhile, histopathological images contain rich and detailed information crucial for disease diagnosis. However, the histopathologica images also encounter several challenges, including high labeling costs, data imbalance, and sensitivity to color variations. To address these issues, we propose a three-view momentum encoder model (View3M) based on multi-view self-supervised learning for histopathological image analysis. First, we employ data augmentation techniques to address the color sensitivity issue in histopathological images. Second, we construct three views to learn image features, and the cosine annealing algorithm is introduced to dynamically update the momentum encoder to maximize the number of negative samples. Next, we design a tri-view consistency contrastive loss function to compare features from different perspectives. Finally, we utilize transfer learning to tackle the challenge of limited sample sizes. Experiments conducted on three publicly available histopathological image datasets, spanning different categories and scales, demonstrate that the model achieves state-of-the-art accuracy and robustness in classification benchmarks.