Knowledge-Based Systems最新文献

{"title":"FACT: Federated Adaptive Cross Training","authors":"Stefan Schrod ,&nbsp;Jonas Lippl ,&nbsp;Andreas Schäfer ,&nbsp;Michael Altenbuchinger","doi":"10.1016/j.knosys.2025.113655","DOIUrl":"10.1016/j.knosys.2025.113655","url":null,"abstract":"<div><div>Federated Learning (FL) facilitates distributed model development to aggregate multiple confidential data sources. The information transfer among clients can be compromised by distributional differences, i.e., by non-i.i.d. data. A particularly challenging scenario is the federated model adaptation to a target client without access to annotated data. We propose Federated Adaptive Cross Training (FACT), which uses the implicit domain differences between source clients to identify domain shifts in the target domain. In each round of FL, FACT cross initializes a pair of source clients to generate domain specialized representations which are then used to enforce a domain invariant data representation. We empirically show that FACT outperforms both state-of-the-art federated and non-federated models on three popular multi-source–single-target benchmarks, and achieves highly competitive performance on single-source–single-target experiments. We further study FACT’s behavior with respect to communication restrictions and the number of participating clients.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"320 ","pages":"Article 113655"},"PeriodicalIF":7.2,"publicationDate":"2025-05-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143942768","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Dynamic graph representation learning via edge temporal states modeling and structure-reinforced transformer","authors":"Shengxiang Hu ,&nbsp;Guobing Zou ,&nbsp;Song Yang ,&nbsp;Shiyi Lin ,&nbsp;Yanglan Gan ,&nbsp;Bofeng Zhang","doi":"10.1016/j.knosys.2025.113661","DOIUrl":"10.1016/j.knosys.2025.113661","url":null,"abstract":"<div><div>The rapid proliferation of time-evolving networks has rendered dynamic graph representation learning increasingly crucial for real-world applications, as existing approaches that combine recurrent neural networks (RNNs) with graph neural networks (GNNs) face two critical limitations: insufficient modeling of edge temporal states and their impact on node feature evolution, along with the inherent over-smoothing problem of GNNs that impedes effective extraction of global structural features. To address these challenges, we introduce the <u><strong>R</strong></u>ecurrent <u><strong>S</strong></u>tructure-reinforced <u><strong>G</strong></u>raph <u><strong>T</strong></u>ransformer (RSGT), a novel framework that advances dynamic graph representation learning through two key innovations. First, it introduces a principled approach to explicitly model edge temporal states using differentiated edge types and weights derived from sequential snapshot analysis, effectively integrating temporal dynamics into the graph’s topological structure. Second, it designs a structure-reinforced graph transformer that leverages a recurrent learning paradigm to capture comprehensive node representations, simultaneously encoding both local connectivity patterns and global structural features while preserving temporal evolution characteristics. Comprehensive experiments on four real-world datasets demonstrate RSGT’s superior performance in discrete dynamic graph representation learning, consistently outperforming existing methods in dynamic link prediction tasks.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"320 ","pages":"Article 113661"},"PeriodicalIF":7.2,"publicationDate":"2025-05-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143937822","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"RG4LDL: Renormalization group for label distribution learning","authors":"Chao Tan ,&nbsp;Sheng Chen ,&nbsp;Jiaxi Zhang ,&nbsp;Zilong Xu ,&nbsp;Xin Geng ,&nbsp;Genlin Ji","doi":"10.1016/j.knosys.2025.113666","DOIUrl":"10.1016/j.knosys.2025.113666","url":null,"abstract":"<div><div>Label distribution learning (LDL) is an effective paradigm to address label ambiguity by modeling the relevance of multiple labels to an instance. However, existing LDL methods suffer from challenges such as high model complexity, slow convergence, and limited availability of label distribution-annotated training data. To tackle these issues, we propose RG4LDL, a novel framework that integrates the renormalization group (RG) principle with LDL for the first time. RG4LDL employs a restricted Boltzmann machine (RBM)-based neural network to iteratively extract relevant degrees of freedom, thereby optimizing feature learning and improving predictive accuracy. By combining unsupervised RG learning and supervised LDL prediction in an end-to-end manner, RG4LDL achieves both efficiency and effectiveness. Experimental results on 13 real-world datasets and a synthetic toy dataset demonstrate that RG4LDL significantly outperforms state-of-the-art LDL methods in terms of predictive accuracy and computational efficiency. These results highlight the potential of RG4LDL as a benchmark solution for label distribution learning tasks.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"320 ","pages":"Article 113666"},"PeriodicalIF":7.2,"publicationDate":"2025-05-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143937840","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multi-Level Skeleton Self-Supervised Learning: Enhancing 3D action representation learning with Large Multimodal Models","authors":"Tian He ,&nbsp;Yang Chen ,&nbsp;Xu Gao ,&nbsp;Ling Wang ,&nbsp;Rui Huang ,&nbsp;Hong Cheng","doi":"10.1016/j.knosys.2025.113660","DOIUrl":"10.1016/j.knosys.2025.113660","url":null,"abstract":"<div><div>Self-Supervised Learning (SSL) has proven effective in skeleton-based action understanding, drawing increasing research attention. Previous studies mainly focus on capturing the relationship between joints and skeleton sequences through joint-level masked motion modeling and sequence-level contrastive learning. However, these methods overlook subtle semantic connections between similar movements, leading to poor feature discrimination of such actions and impacting downstream task performance. In this paper, we propose a Multi-Level Skeleton Self-Supervised Learning (MLS<span><math><msup><mrow></mrow><mrow><mn>3</mn></mrow></msup></math></span>L) framework that integrates joint, sequence, and semantic-level SSL in a complementary manner for fine-grained action understanding. Specifically, We first design topology-based mask reconstruction for joint-level SSL and tempo-independent contrastive learning for sequence-level SSL. For semantic-level SSL, we leverage pre-trained Large Multimodal Models (LMMs) to generate discriminative text descriptions for action sequences. Then, we design a weighted soft alignment algorithm to align text descriptions with the corresponding skeletons. This semantic-level representation distillation significantly enhances the ability to distinguish between similar actions. Furthermore, we propose a multi-level collaboration strategy to enable SSL tasks at different levels to jointly learn versatile representations of various granularity, leading to improved learning of action representation features. Our method demonstrates exceptional performance on various downstream tasks, validated on NTU RGB+D, NTU RGB+D 120, and PKUMMD datasets.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"320 ","pages":"Article 113660"},"PeriodicalIF":7.2,"publicationDate":"2025-05-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143942742","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhancing quantum support vector machine for healthcare applications using custom feature maps","authors":"Riya Bansal ,&nbsp;Nikhil Kumar Rajput ,&nbsp;Megha Khanna","doi":"10.1016/j.knosys.2025.113669","DOIUrl":"10.1016/j.knosys.2025.113669","url":null,"abstract":"<div><div>Quantum support vector machine (QSVM), based on the principles of quantum mechanics has revolutionized complex data processing tasks in several healthcare applications. Feature maps play a crucial role in transforming input data into a higher-dimensional space, enabling QSVM to capture intricate patterns and improve classification performance. This study intends to further enhance the performance of the QSVM by introducing five new custom feature maps. Furthermore, the study assesses the performance of these enhancements to the QSVM by empirically validating it for classification on four medical open-source datasets. The performance of QSVM using the custom feature maps is also compared with two standard feature maps (ZFeatureMap and ZZFeatureMap) available in Qiskit framework. The results indicate that custom feature maps outperform standard ones with an increase of up to 5% in Area Under Receiver Operating Characteristic Curve (AUC) values, 18% in F1-score, and 25% in Matthews Correlation Coefficient (MCC) values.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"320 ","pages":"Article 113669"},"PeriodicalIF":7.2,"publicationDate":"2025-05-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143937838","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Granularity Fusion Transformer: Learning multi-granularity patterns for time-series forecasting","authors":"Jinwoo Park ,&nbsp;Hyeongwon Kang ,&nbsp;Seunghun Han ,&nbsp;Pilsung Kang","doi":"10.1016/j.knosys.2025.113644","DOIUrl":"10.1016/j.knosys.2025.113644","url":null,"abstract":"<div><div>Time series data consist of continuous observations collected over time. Theoretically, time series data form a continuous trajectory, but in practice, the actual data used in the real-world are discrete time series obtained by sampling from a continuous trajectory. By considering the sampling interval, it is possible to collect data with different levels of information, referred to as granularity in time series data. Most existing studies assume single-granularity, which leads to failure in capturing variability occurring at different levels of granularity. To address these issues, we propose the Granularity Fusion Transformer (GFT), which addresses characteristics at different granularity levels. GFT estimates multi-granularity data by simultaneously considering the information of period and phase at the single-granularity level. To merge granularities that are segregated into different levels, we propose a patch-wise cross-attention based Granularity Fusion Encoder. Extensive experiments on six datasets demonstrate that the proposed method outperformed benchmark models by reducing MSE by 35.27<span><math><mtext>%</mtext></math></span> and MAE by 22.80<span><math><mtext>%</mtext></math></span>, thereby achieving more accurate predictions closer to the actual values. These results highlight the usefulness of multi-granularity in time series forecasting tasks.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"320 ","pages":"Article 113644"},"PeriodicalIF":7.2,"publicationDate":"2025-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143942745","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A survey of language-grounded multimodal 3D scene understanding","authors":"Ruilong Ren ,&nbsp;Xinyu Zhao ,&nbsp;Weichen Xu ,&nbsp;Jian Cao ,&nbsp;Xinxin Xu ,&nbsp;Xing Zhang","doi":"10.1016/j.knosys.2025.113650","DOIUrl":"10.1016/j.knosys.2025.113650","url":null,"abstract":"<div><div>As an emerging task bridging vision and language, Language-grounded Multimodal 3D Scene Understanding (3D-LMSU) has attracted significant interest across various domains, such as robot navigation and human–computer interaction. It aims to generate detailed and precise responses to textual queries related to 3D scenes. Despite the popularity and effectiveness of existing methods, the absence of a comprehensive survey hampers further development. In this study, we present the first systematic survey of recent progress in addressing this gap. We start with a concise overview of the background, including the problem definition and available benchmark datasets. Subsequently, we introduce a novel taxonomy that provides a comprehensive classification of existing methods based on technologies and tasks. We then present the evaluation metrics for each task, along with the performance results of various methods. Furthermore, we offer insightful discussions from three critical perspectives: data, framework, and training. Finally, we conclude the paper by highlighting several promising avenues for future research. This study synthesizes the field and guides researchers toward further exploration.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"321 ","pages":"Article 113650"},"PeriodicalIF":7.2,"publicationDate":"2025-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144088711","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A reinforcement learning-based dynamic multi-objective optimization approach for integrated timetabling and vehicle scheduling","authors":"Yindong Shen ,&nbsp;Wenliang Xie","doi":"10.1016/j.knosys.2025.113735","DOIUrl":"10.1016/j.knosys.2025.113735","url":null,"abstract":"<div><div>Dynamic integrated timetabling and vehicle scheduling (D-ITVS) is essential for mitigating the negative impacts of service disruptions. It involves multiple rescheduling stages, with inherent optimization similarities across these stages. However, existing optimization approaches for the <span>D-ITVS</span> problem have not systematically exploited these similarities, overlooking the potential for decision knowledge from previous stages to inform the current stage. To address this gap, this paper proposes a reinforcement learning-based dynamic multi-objective optimization approach (RL-DMOA), which focuses on transferring decision knowledge between rescheduling stages. This approach models the optimization process of each rescheduling stage in the <span>D-ITVS</span> problem as a Markov decision process, incorporating a state space with vehicle information, action space for vehicle assignment, and a multi-objective reward function. A multi-objective deep reinforcement learning (M-DRL) agent is employed within the RL-DMOA to select actions based on the state at each decision point. The agent is constructed on a multi-objective deep Q-learning network (M-DQN), with a Q-value adjustment layer incorporated to prevent the selection of invalid actions. To select optimal actions while balancing the conflicts among multiple objectives, the M-DRL agent applies a non-dominated sorting selection strategy. Experimental results demonstrate that the proposed RL-DMOA is capable of generating timetables and vehicle schedules with reduced costs, enhanced robustness, and improved convergence and diversity across all rescheduling stages. By balancing operational costs and passenger service quality, these improvements benefit transit operators, and during daily operations, passengers enjoy reduced travel costs and enhanced service reliability.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"321 ","pages":"Article 113735"},"PeriodicalIF":7.2,"publicationDate":"2025-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144072287","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Causal representation learning in offline visual reinforcement learning","authors":"Yaru Zhang,&nbsp;Kaizhou Chen,&nbsp;Yunlong Liu","doi":"10.1016/j.knosys.2025.113565","DOIUrl":"10.1016/j.knosys.2025.113565","url":null,"abstract":"<div><div>Real-world reinforcement learning (RL) applications contend with high-dimensional visual observations contaminated by confounding factors, which induce spurious correlations and obscure decision-relevant information. Compounding this issue, the inability to interact online necessitates reliance on pre-collected datasets, thereby hampering a deeper understanding of complex environment structures. In this work, by focusing on the causal rather than spurious correlations in the input and explicitly distinguishing between task-related and task-irrelevant elements of the causal variables, we propose a mask-based algorithm for learning task-related minimal causal state representations, namely MMCS. Specifically, MMCS guides the decoupling of minimal causal variables through mask network partitioning and jointly enforcing conditional independence and causal sufficiency, thereby eliminating unnecessary dependencies between variables and uncovering causal dependency structures. More importantly, MMCS is decoupled from downstream policy learning, and can function as a plug-in method compatible with any offline reinforcement learning algorithm. Empirical results on the Visual-D4RL benchmark demonstrate that MMCS significantly improves performance and sample efficiency in downstream policy learning. In addition, its robust performance in various distraction environments highlights the potential of MMCS to improve the generalizability of offline RL, especially under conditions of limited data and visual distractions. Code is available at <span><span>https://github.com/DMU-XMU/MMCS.git</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"320 ","pages":"Article 113565"},"PeriodicalIF":7.2,"publicationDate":"2025-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143931696","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"KALD: A Knowledge Augmented multi-contrastive learning model for low resource abusive Language Detection","authors":"Rui Song ,&nbsp;Fausto Giunchiglia ,&nbsp;Yingji Li ,&nbsp;Jian Li ,&nbsp;Jingwen Wang ,&nbsp;Hao Xu","doi":"10.1016/j.knosys.2025.113619","DOIUrl":"10.1016/j.knosys.2025.113619","url":null,"abstract":"<div><div><em>Warning: This paper contains insulting statements that may cause discomfort for readers.</em></div><div>With the development of online social media, quite a few methods focus on automatic Abusive Language Detection (ALD), which requires numerous annotations as the basis for reliable classifier training. However, the labor-intensive, expensive, and time-consuming data labeling process brings difficulties to the acquisition of the annotations. Although some studies have improved the model performance in the absence of labeled data by studying cross-domain generalization and semi-supervised learning, there is still a lack of specific research on making full use of prior knowledge to improve detection effectiveness in the context of limited resources. To solve this problem, we propose a <strong>K</strong>nowledge <strong>A</strong>ugmented abusive <strong>L</strong>anguage <strong>D</strong>etection framework (KALD), to fully utilize three kinds of prior knowledge: lexical knowledge, sample knowledge, and category knowledge. First, lexicon knowledge is injected into the language model to promote its focus on abusive keyword by context reconstruction. Meanwhile Lexicon-based data augmentation is used to obtain reasonable positive samples necessary for contrastive learning. Subsequently Joint optimization of multi-contrastive learning is applied to encourage language models to learn stable sample-level and in-class representations. The following tasks are performed on the four public datasets to verify the validity of the proposed method (a) ALD (b) semi-supervised ALD And (c) cross-domain abusive language generalization. For semi-supervised ALD, the proposed framework has an average improvement of 2.19% with different sample size settings compared to the most advanced baseline approach and 3.58% compared to the basic language model. For cross-domain abusive language generalization, the proposed framework has an average improvement of 2.58% and 3.42% compared with the most advanced baseline approach and the basic language model, separately.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"321 ","pages":"Article 113619"},"PeriodicalIF":7.2,"publicationDate":"2025-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144068744","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}