Prototype-based personalized federated learning methods have emerged as a promising strategy due to their ability to represent client-specific class characteristics effectively through learned class prototypes. These prototypes capture salient features of client-local data, facilitating personalized model adaptation. However, existing prototype-based aggregation strategies predominantly rely on weighted averaging, implicitly assuming prototype consistency across clients. This assumption neglects the intrinsic heterogeneity and non-independent and identically distributed (non-IID) nature of client data, compelling diverse local prototypes to align toward a singular global prototype and consequently causing significant aggregation bias. Motivated by observations from intra-class feature saliency analysis, we identify that clients inherently emphasize distinct feature regions even for the same class. To leverage this intra-class diversity, we introduce FedIC, a novel prototype clustering and collaborative classifier optimization approach. Specifically, FedIC first clusters prototypes based on intra-class similarity to form intra-class prototype subspaces, ensuring that aggregation occurs exclusively within each cluster, thus eliminating the bias stemming from forced global unification. To further exploit the benefits of intra-cluster collaboration, we quantify the combined predictive gains of classifiers from clients within the same cluster as a function of classifier combination weights. This targeted aggregation and collaborative optimization strategy effectively circumvents the bias introduced by global alignment. Extensive experiments under various non-IID settings show that FedIC significantly outperforms existing Prototype-based and Clustered PFL Methods.

The limited availability of high-quality training data poses a persistent challenge for synthetic aperture radar (SAR) target classification. Existing data augmentation methods mainly adopt a simplistic application of GAN-based style transfer techniques to directly synthesize pseudo-SAR images from optical images. However, our in-depth analysis of this cross-modal conversion reveals that such straightforward strategies primarily focus on transferring high-level semantic information (e.g., target shapes), thus failing to adequately capture the essential low-level features unique to SAR imagery (e.g., scattering textures). To address this inherent trade-off between high-level semantic preservation and low-level feature authenticity, we propose a Hierarchical Feature-Constrained GAN (HiFC-GAN) tailored for optical-to-SAR style transfer. Specifically, HiFC-GAN enhances the representation of low-level SAR features by introducing local texture contrast constraints at shallow layers, while introducing explicit feature mapping constraints at deeper layers to maintain high-level semantic consistency throughout the reconstruction process. Experimental results demonstrate that HiFC-GAN significantly outperforms existing GAN-based techniques in image generation quality, particularly improving the low-level feature authenticity of pseudo-SAR images. Moreover, the generated pseudo-SAR images further improve the performance of downstream target classification tasks, yielding accuracy gains ranging from 3.56% to 5.90% on average with mainstream CNN-based models.

Entity alignment (EA) is crucial for integrating knowledge graphs (KGs) constructed from diverse sources. Conventional unsupervised EA approaches attempt to eliminate human intervention but often suffer from accuracy limitations. With the rise of large language models (LLMs), leveraging their capabilities for EA presents a promising direction. However, it introduces new challenges: formulating the LLM-based EA problem and extracting the background knowledge in LLMs to realize EA without human intervention. This paper proposes HLMEA, a novel hybrid language model-based unsupervised EA method. HLMEA formulates the EA task into a filtering and single-choice problem and synergistically integrates small language models (SLMs) and LLMs. Specifically, SLMs filter candidate entities based on textual representations generated from KG triples. Then, LLMs refine this selection to identify the most semantically aligned entities. An iterative self-training mechanism allows SLMs to distill knowledge from LLM outputs, enhancing the EA ability of hybrid language models in subsequent rounds cooperatively. We also conducted extensive experiments on benchmark datasets to evaluate HLMEA's performance. The results demonstrate that HLMEA significantly outperforms unsupervised and even supervised EA baselines, proving its potential for scalable and effective EA across large KGs. The code and data are available at \url{https://github.com/xnjin-ai/HLMEA}.

Negative transfer (NF) is a critical challenge in personalized federated learning (pFL). Existing methods primarily focus on adapting local data distribution on the client side, which can only resist NF, rather than avoid NF itself. To tackle NF at its root, we investigate its mechanism through the lens of the global model, and argue that it is caused by update conflicts among clients during server aggregation. In light of this, we propose a conflict-free client update aggregation strategy (ConFREE), which enables us to avoid NF in pFL. Specifically, ConFREE guides the global update direction by constructing a conflict-free guidance vector through projection and utilizes the optimal local improvements of the worst-performing clients near the guidance vector to regularize server aggregation. This prevents the conflicting components of updates from transferring, achieving balanced updates across different clients. Notably, ConFREE is model-agnostic and can be straightforwardly adopted as a complement to enhance various existing NF-resistance methods implemented on the client side. Extensive experiments demonstrate substantial improvements to existing pFL algorithms by leveraging ConFREE.

Temporal Knowledge Graphs (TKGs) serve as crucial tools for representing dynamic changes in the real world. Extrapolation reasoning within TKGs aims to predict entirely unknown future facts based on limited historical data, offering considerable practical value across various fields. However, existing methods generally focus on the recurrence and periodicity of historical facts, while overlooking the dynamic interactions associated with future facts. Moreover, these methods fail to capture historical evolutionary patterns, which grow increasingly complex as historical data accumulates. To this end, we propose TCPN, a novel Temporal Pyramidal Recurrent Network with contrastive learning for TKG extrapolation reasoning. Specifically, TCPN leverages a temporal pyramidal recurrent network to capture historical dependencies across multiple temporal scales, thereby refining temporal feature representations over extended time spans. Furthermore, TCPN seamlessly integrates contrastive learning to effectively align historical information with query semantics relevant to future facts. Lastly, we incorporate an adaptive time-aware mechanism, which uniformly models long-short term dependencies in time series with different granularities, explicitly fusing temporal feature information. Extensive experiments on four widely used TKG datasets show that TCPN significantly outperforms state-of-the-art methods across all metrics.

Server aggregation conflict is a key challenge in personalized federated learning (PFL). While existing PFL methods have achieved significant progress with shallow base models (e.g., four-layer CNNs), they often overlook the negative impacts of deeper base models on personalization mechanisms. In this paper, we identify the phenomenon of deep model degradation in PFL, where as base model depth increases, the model becomes more sensitive to local client data distributions, thereby exacerbating server aggregation conflicts and ultimately reducing overall model performance. Moreover, we show that these conflicts manifest in insufficient global average updates and mutual constraints between clients. Motivated by our analysis, we proposed a two-stage conflict-aware layer-wise mitigation algorithm (FedCALM), which first constructs a conflict-free global update to alleviate negative conflicts, and then maximizes the benefits of all clients through a conflict-aware strategy. Notably, our method naturally leads to a selective mechanism that balances the tradeoff between clients involved in aggregation and the tolerance for conflicts. Consequently, it can boost the positive contribution to the clients even with the greatest conflicts with the global update. Extensive experiments across multiple datasets and deeper base models demonstrate that FedCALM outperforms four state-of-the-art (SOTA) methods by up to 9.88% and seamlessly integrates into existing PFL methods with performance improvements of up to 9.01%.

Extrapolation reasoning on Temporal Knowledge Graphs (TKGs) plays a pivotal role in various systems, including retrieval, recommendation, and Q&A. Traditional TKG reasoning methods tend to emphasize modeling the local and global features of facts, often overlooking the alignment with query semantics. Crucially, these methods are challenging in generating explicit reasoning paths. To address these gaps, we propose an innovative framework (LogiQ) for extrapolation reasoning on TKGs, steered by temporal logic rules and queries. Specifically, LogiQ incorporates temporal logic rules and implements a rule-guided reward mechanism, directing Reinforcement Learning (RL) agents toward actions more aligned with rules. Additionally, LogiQ merges temporal queries with neighbor aggregation, ensuring that candidate actions not only encapsulate neighboring factual data but also embody query semantics. This dual focus enables LogiQ to guide actions to find reasoning paths in limited steps strategically. Extensive experiments on four real-world TKG datasets demonstrate the superior performance of LogiQ across all metrics compared to existing state-of-the-art models.

Session-based recommendation (SBR) provides personalized recommendations by analyzing the interactions of anonymous session users. Recent approaches based on graph neural networks (GNNs) focus on pairwise relations to infer potential user preferences. However, real-world user interactions are often complex, involving high-order relations that GNNs may not fully capture. Hypergraphs can naturally model high-order relations, while having untapped potential in SBR. In this paper, we propose a hypergraph convolutional network (MoHyNet) based on hypergraph motifs to improve SBR. Specifically, we construct a hypergraph convolution to extract high-order relations among users, thereby reducing the impact of intra-session irrelevant information on recommendations. Additionally, we introduce hypergraph motifs to characterize users’ behavioral patterns, thus enhancing inter-session information mining. Besides, we incorporate contrastive learning to strengthen the representation of the current session. Extensive experiments on multiple real-world datasets demonstrate the superiority of our proposed model over state-of-the-art approaches.

Many existing personalized federated learning (PFL) methods utilize clustering-based aggregation to group clients with similar data characteristics, improving model performance by promoting collaboration among clients with shared features. While this method effectively mitigates some challenges posed by data heterogeneity, it predominantly relies on static data features, making it challenging to capture the dynamic changes in client models during iterative training. This limitation impedes accurate clustering based on evolving model updates. To address this issue, we propose a Gradient-Driven Adaptive Clustering method in PFL (GradPFL), which more effectively captures the personalized deviations in locally updated models. Our approach also introduces an adaptive historical gradient mechanism that refines the clustering process by incorporating both current and past update characteristics. This enables more accurate model aggregation that adapts to ongoing changes in client models during training. Experimental results demonstrate that GradPFL outperforms existing clustering-based PFL methods, especially in more complex non-IID environments.

Graph Edit Distance (GED) is a classical graph similarity metric. Since exact GED computation is NP-hard, existing GNN-based methods try to approximate GED in polynomial time. However, they still lack support for edge labels or the ability to generate an edit path. To address these limitations, we propose a hybrid method named CGEDN based on Graph Neural Networks and learnable node matching. Specifically, CGEDN starts with cross-graph feature aggregation layers, which generate node embeddings with fine-grained interaction features and support edge labels as well. Next, two node matching pipelines are applied to obtain a node matching confidence matrix and a cost matrix, where the confidence matrix is supervised by optimal node matchings corresponding to GED. Finally, we calculate the weighted sum of the cost matrix with a bias value to regress GED only, or search for the top-k most promising node matchings based on confidence matrix to approximate GED and recover an edit path simultaneously. The experimental results on real and synthetic graphs demonstrate that CGEDN significantly outperforms the best result of existing approximate methods.

Existing SAR image classification methods based on Contrastive Learning often rely on sample generation strategies designed for optical images, failing to capture the distinct semantic and physical characteristics of SAR data. To address this, we propose Physics-Driven Contrastive Mutual Learning for SAR Classification (PCM-SAR), which incorporates domain-specific physical insights to improve sample generation and feature extraction. PCM-SAR utilizes the gray-level co-occurrence matrix (GLCM) to simulate realistic noise patterns and applies semantic detection for unsupervised local sampling, ensuring generated samples accurately reflect SAR imaging properties. Additionally, a multi-level feature fusion mechanism based on mutual learning enables collaborative refinement of feature representations. Notably, PCM-SAR significantly enhances smaller models by refining SAR feature representations, compensating for their limited capacity. Experimental results show that PCM-SAR consistently outperforms SOTA methods across diverse datasets and SAR classification tasks.

Class imbalance with node quantity or node topology challenges graph node classification in real-world, such as fake user identification and fraud finance detection in social networks. Current studies mitigate the detrimental effects of class imbalance on graph neural networks through data augmentation and weight adjustment. However, these methods generate unreliable data and lack global generalizability, which results in poor classification performance for class-imbalanced nodes. To tackle the above issues, we propose a Graph Diffusion Enhanced Hypergraph method (GraphDEH), which introduces a hypergraph to capture high-order data correlations, strengthening semantic relational space of nodes. Specifically, we design a generator to constructs reliable hyperedges, which mitigate insufficient learning of minority classes caused by node quantity imbalance. Furthermore, we design an enhancer that employs graph diffusion to evaluate global influence among nodes, which increases homogeneous information in neighborhoods of minority-class nodes for node topology imbalance. GraphDEH outperforms multiple benchmark methods in class-imbalanced node classification across four benchmark datasets.

User requests in edge computing environments are inherently decentralized and dynamic, posing significant challenges for efficient and adaptive service replica placement. To address this, we formulate the service replica placement problem in an edge-cloud collaborative environment, explicitly incorporating the spatio-temporal distribution of user requests. By capturing spatial and temporal correlations, we predict future request patterns to enable forward-looking replica placement. Given the NP-hard nature of the optimization problem, we design a DRL algorithm that optimizes replica placement decisions based on predictive modeling. To validate our approach, we conduct extensive experiments on real-world datasets across two typical application scenarios―grid-based and graph-based request distributions. Experimental results show our method reduces average response latency by up to 59.6% and boosts service provider profitability by 4.85% compared to reactive and temporal-only baselines. The proposed framework provides a novel and effective solution for proactive service provisioning in edge computing environments.

Most existing synthetic aperture radar (SAR) ship classification technologies heavily rely on correctly labeled data, ignoring the discriminate features of unlabeled SAR ship images. Even though researchers try to enrich CNN-based features by introducing traditional handcrafted features, existing methods easily cause information redundancy and fail to capture the interaction between them. To address these issues, we propose a novel dual-stream contrastive predictive network (DCPNet), which consists of two asymmetric tasks and a false negative sample elimination module. The first task is to construct positive sample pairs, guiding the core encoder to learn more general representations. The second task is to encourage adaptive capture of the correspondence between deep features and handcrafted features, achieving knowledge transfer within the model, and effectively improving the redundancy caused by the feature fusion. To increase the separability between clusters, we also design a cluster-level task. The experimental results on OpenSARShip and FUSAR-Ship datasets demonstrate the improvement in classification accuracy of supervised models and confirm the capability of learning effective representations of DCPNet.

In current synthetic aperture radar (SAR) object classification, one of the major challenges is the severe overfitting issue due to the limited dataset (few-shot) and noisy data. Considering the advantages of knowledge distillation as a learned label smoothing regularization, this paper proposes a novel Double Reverse Regularization Network based on Self-Knowledge Distillation (DRRNet-SKD). Specifically, through exploring the effect of distillation weight on the process of distillation, we are inspired to adopt the double reverse thought to implement an effective regularization network by combining offline and online distillation in a complementary way. Then, the Adaptive Weight Assignment (AWA) module is designed to adaptively assign two reverse-changing weights based on the network performance, allowing the student network to better benefit from both teachers. The experimental results on OpenSARShip and FUSAR-Ship demonstrate that DRRNet-SKD exhibits remarkable performance improvement on classical CNNs, outperforming state-of-the-art self-knowledge distillation methods.

Tingxuan Chen, Jun Long, Liu Yang, Zidong Wang, Yongheng Wang, Xiongnan Jin. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Temporal Knowledge Graphs (TKGs) serve as indispensable tools for dynamic facts storage and reasoning. However, predicting future facts in TKGs presents a formidable challenge due to the unknowable nature of future facts. Existing temporal reasoning models depend on fact recurrence and periodicity, leading to information degradation over prolonged temporal evolution. In particular, the occurrence of one fact may influence the likelihood of another. To this end, we propose THCN, a novel Temporal Causal Convolutional Network based on Hawkes processes, designed for temporal reasoning under the extrapolation setting. Specifically, THCN harnesses a temporal causal convolutional network with dilated factors to capture historical dependencies among facts spanning diverse time intervals. Then, we construct a conditional intensity function based on Hawkes processes for fitting the likelihood of fact occurrence. Importantly, THCN pioneers a dual-level dynamic modeling mechanism, enabling the simultaneous capture of the collective features of nodes and the individual characteristics of facts. Extensive experiments on six real-world TKG datasets demonstrate our method significantly outperforms the state-of-the-art across all four evaluation metrics, indicating that THCN is more applicable for extrapolation reasoning in TKGs.

Action segmentation plays an important role in video understanding, which is implemented by frame-wise action classification. Recent works on action segmentation capture long-term dependencies by increasing temporal convolution layers in Temporal Convolution Networks (TCNs). However, high layers in TCNs are more coarse access to video features, resulting in the loss of fine-grained information for frame-wise action classification. To address the above issues, we propose a novel Attention-based Temporal Convolution (ATC) block to capture fine-grained information of temporal dependencies for frame-wise action classification by self-attention mechanism. Via stacking ATC blocks, we design a Stacking-based Attention Temporal Convolutional Network (SATC) to adaptively capture long-term and short-term dependencies, according to the semantic similarity of features on different temporal receptive fields simultaneously. The experimental results demonstrate that our SATC outperforms other baselines on all three challenging datasets: GTEA, 50Salads and Breakfast.

Multimodal learning among video and audio has shown significant performance improvement in violence detection. However, video and audio do not contribute consistently, and the video modality tends to dominate when determining whether a certain scene contains violent events. In fact, a few recent multimodal learning methods for violence detection do not fully consider data differences between various modalities, which lead to optimization imbalance problem during training, and ultimately result in insufficient performance. To address this issue, we propose a Multimodal Contrastive Learning (MCL) method to make full use of video and audio information for violence detection. In specific, to avoid the video modality dominating the model training, we design a multi-encoder framework to perform task-driven feature encoding on video and audio respectively. To reduce information loss during multimodal fusion, we introduce a contrastive learning task to capture semantically consistent representations. We conduct extensive experiments on XD-Violence dataset, showing that our proposed MCL achieves an average precision improvement of 2.34% against the state-of-the-art baseline.

Multi-feature SAR ship classification aims to build models that can process, correlate, and fuse information from both handcrafted and deep features. Although handcrafted features provide rich expert knowledge, current fusion methods inadequately explore the relatively significant role of handcrafted features in conjunction with deep features, the imbalances in feature contributions, and the cooperative ways in which features learn. In this paper, we propose a novel multi-feature collaborative fusion network with deep supervision (MFCFNet) to effectively fuse handcrafted features and deep features for SAR ship classification tasks. Specifically, our framework mainly includes two types of feature extraction branches, a knowledge supervision and collaboration module, and a feature fusion and contribution assignment module. The former module improves the quality of the feature maps learned by each branch through auxiliary feature supervision and introduces a synergy loss to facilitate the interaction of information between deep features and handcrafted features. The latter module utilizes an attention mechanism to adaptively balance the importance among various features and assign the corresponding feature contributions to the total loss function based on the generated feature weights. We conducted extensive experimental and ablation studies on two public datasets, OpenSARShip-1.0 and FUSAR-Ship, and the results show that MFCFNet is effective and outperforms single deep feature and multi-feature models based on previous internal FC layer and terminal FC layer fusion. Furthermore, our proposed MFCFNet exhibits better performance than the current state-of-the-art methods.

Graph-clustering algorithms are designed to split large-scale Resource Description Framework (RDF) graphs into subgraphs to improve RDF query performance. However, triple semantics and graph structures embedded in RDF graphs are often ignored during the RDF graph partition. Accordingly, this study utilizes the Leiden algorithm for uncovering community structure to cluster RDF graphs. We propose an optimized WLeidenRDF algorithm to uncover RDF communities and cluster vertex with strong semantics. Different weights are set to predicates in RDF triples to identify semantic-relevance degree and ensure semantic connections after RDF clustering and segmentation. The experiments on WatDiv data sets demonstrate that our WLeidenRDF algorithm obtains better partitions in accordance with modularity than those of the Leiden algorithm. We implement our algorithm on Presto distributed SQL query engine. Experimental results indicate that our algorithm can substantially reduce query time by clustering RDF graphs than with other RDF query methods.

The tags of open-source software (OSS) are important for managing and retrieving a massive amount of OSS in the OSS community, untagged OSS makes managing and retrieving OSS on GitHub difficult. However, developers sometimes neglect to write tag for repositories. For example, in our collected dataset with over 43K GitHub repositories, more than 32 % of the repository are unlabeled. To alleviate this problem, we propose an approach to automatically generate repository tag based on a neural network and LDA by utilizing package dependencies and readme among OSS in communities. We design an algorithm for extracting the tag features of dependent OSS packages and build dependent feature vectors for OSS. We then combine the vectors with topic of OSS readme file as input to train the neural network and obtain the tag distribution probability of OSS, and subsequently, recommend tags for OSS. Experiments are performed on the OSS dataset that we collected from GitHub, over 43K repositories and evaluate our approach on this dataset. Experiment results show that DepTagRec performs better than other methods in terms of precision and recall, particularly on recall when recommending the top 10 tags for OSS.

Selecting professional and authoritative experts to evaluate projects is an essential process in order to assure the quality of projects. In this paper, we present a semantic-based expert information system, which search for expert information based on semantic and knowledge reasoning, and rank the search results according to the scientific capability of experts. We design software architecture for semantic-based information service system (Esoogle). Expert information ontology is defined to store the information of experts, and reasoning rules are defined for semantic-based reasoning. Assessment model based on TOPSIS is built to estimate the scientific capability for ranking. Applications show Esoogle improves the recall and precision of semantic-based searching service compared to the traditional systems and ranking service is helpful for the users to select suitable experts quickly.