Neural Networks最新文献

{"title":"Manod: A multi-modal anomaly detection framework for distributed system","authors":"Wen Liu ,&nbsp;Degang Sun ,&nbsp;Haitian Yang ,&nbsp;Yan Wang ,&nbsp;Weiqing Huang","doi":"10.1016/j.neunet.2025.107999","DOIUrl":"10.1016/j.neunet.2025.107999","url":null,"abstract":"<div><div>Distributed infrastructure has been widely deployed in large-scale software systems in recent years to meet the growing demand for applications, due to its scalability and resource-sharing characteristics. Accurately predicting and identifying anomalies is critical to ensure the stable and reliable running of complex distributed systems. System abnormalities can often be reflected through key performance indicators and logs. Metrics provide quantitative measures of system performance and operational status, while logs record various events that occur in the system. Current approaches typically rely on a single data source to detect anomalies, which may lead to false positives and limit the accuracy of failure detection. A combination of these two data modalities can provide a comprehensive view of the system behavior. In this work, we propose a semi-supervised fault detection method, Manod, to monitor the health state of the system based on multimodal data. To obtain the discriminative representations, it employs a graph-based hierarchical encoding approach and leverages pre-trained language models for modeling metrics and logs, respectively. Then, it adopts a novel gated attention fusion method to integrate heterogeneous information. Extensive experiments on two datasets validate the effectiveness of our proposed Manod. It achieves F1-scores of 0.870 and 0.934 on one simulation dataset (D1) and one real-world dataset (D2), respectively, and significantly outperforms all baseline models. This demonstrates its capacity in mitigating both false positives and false negatives.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107999"},"PeriodicalIF":6.3,"publicationDate":"2025-08-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144903303","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":"Large language modeling of hallucinatory problem mitigation based on the wheel of emotions","authors":"Zhenyu Wang ,&nbsp;Jianmin Wang ,&nbsp;Zenan Lu ,&nbsp;Fang You","doi":"10.1016/j.neunet.2025.107996","DOIUrl":"10.1016/j.neunet.2025.107996","url":null,"abstract":"<div><div>Large Language Models (LLMs) have demonstrated remarkable generative capabilities across a wide range of natural language processing tasks. However, the frequent occurrence of hallucinations—outputs that appear plausible but are factually incorrect or logically inconsistent—poses a significant challenge to the reliability and practical utility of these models. This paper proposes a novel Emotion-Augmented Inference (EAI) method based on the Wheel of Emotions, aiming to mitigate hallucinations in multimodal generation tasks involving LLMs. EAI integrates two core mechanisms: visual-contrastive decoding and affective textual symbolization, which jointly enable the perception, regulation, and reconstruction of emotional signals during generation. These mechanisms enhance emotional coherence and semantic reliability in the model's outputs. Experimental results on two multimodal datasets, MSCOCO and GQA, show that EAI achieves improvements of 4%–8 % over baseline models in terms of key metrics such as accuracy, precision, recall, and F1-score. Additionally, under three emotional contexts—neutral (S1), positive (S2), and negative (S3)—EAI demonstrates particularly strong performance in hallucination suppression. In the S3 condition, accuracy improves by 5.48% and 2.23% compared to S1 and S2, respectively. These findings also indicate that EAI enhances the ability to manage emotion and maintain textual coherence. In summary, EAI not only stabilizes hallucination suppression in multimodal generation but also provides a new perspective for interpreting the emotional states embedded in LLM outputs. The proposed method offers a promising direction for building more trustworthy, controllable, and human-centered AI systems.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107996"},"PeriodicalIF":6.3,"publicationDate":"2025-08-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144879975","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 prompt-aware knowledge-tuning framework for histopathology subtype classification with scarce annotation","authors":"Bo Yu ,&nbsp;Jiuman Song ,&nbsp;Lele Cong ,&nbsp;Xianling Cong ,&nbsp;Jouke Dijkstra ,&nbsp;Philip S. Yu ,&nbsp;Hechang Chen","doi":"10.1016/j.neunet.2025.107993","DOIUrl":"10.1016/j.neunet.2025.107993","url":null,"abstract":"<div><div>Artificial intelligence can assist pathologists in diagnosing histopathology subtypes, enabling precision medicine and improving survival rates. Many approaches employ multi-scale models or combine knowledge to implement subtype diagnosis. However, they fail to identify explicit features most relevant to subtypes adaptively, resulting model relying heavily on extensive annotation. Moreover, knowledge is qualitatively represented by coarse-grained methods, such as using 0 or 1 to indicate negative or positive samples. However, they cannot be quantitatively described with a fine-grained process, such as with a probability of 0.23 or 0.81. In this paper, we propose a prompt-aware knowledge-tuning model called PAKT for subtype classification, which provides an adaptive feature generation while representing knowledge quantitatively with scarce annotation. Specifically, we design a prompt-aware module that adaptively predicts multi-scale histological probabilities. The pre-trained encoder can leverage vision prompts to obtain explicit features without extensive annotation. Furthermore, a knowledge-tuning module is constructed to provide sensible diagnostic processes. The trainable weight matrix can quantitatively represent diagnosis knowledge, reflecting the influence of different histological probabilities on subtypes. PAKT performs better than state-of-the-art methods in diagnosing subtypes, achieving an average performance improvement of over 10 %, as evidenced by extensive experimentation on both public and in-house datasets, thus validating its effectiveness. Moreover, its complexity is significantly reduced without losing performance compared with baselines. Code: <span><span>https://github.com/Dennis-YB/PAKT.git</span><svg><path></path></svg></span></div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107993"},"PeriodicalIF":6.3,"publicationDate":"2025-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144879969","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-site brain disease identification based on tensor decomposition and personalized federated learning","authors":"Chaojun Zhang ,&nbsp;Jing Yang ,&nbsp;Yuan Gao ,&nbsp;Xiangli Yang ,&nbsp;Shaojun Zou ,&nbsp;Jieming Yang","doi":"10.1016/j.neunet.2025.107987","DOIUrl":"10.1016/j.neunet.2025.107987","url":null,"abstract":"<div><div>Brain diseases significantly impact physical and mental health, making the development of models to identify biomarkers for early diagnosis essential. However, building high-quality models typically relies on large-scale datasets, while the privacy-sensitive nature of medical data often restricts its sharing and utilization. Multi-site studies provide a potential solution by integrating data from various sources, yet existing methods frequently neglect site-specific private features, such as demographic information. Therefore, in this paper, we propose a simple yet effective framework based on Tensor Decomposition and Personalized Federated Learning (TDPFL) for multi-site brain disease recognition, while protecting these private features. On the central server, we designed a dual feature aggregation module to facilitate efficient knowledge sharing among sites. On the client side, we introduced a personalized branch to safeguard private information (<em>i.e.</em>, age, gender, and education) and developed a tensor decomposition module to extract features from subjects’ brain scan data. Furthermore, we developed a dynamic prototype aggregation module to monitor evolving brain features over time. This mechanism enhances the model’s capacity to capture these dynamics, thereby improving classification and prediction accuracy. Experiments on two publicly available rs-fMRI datasets across six sites showed that TDPFL outperformed baseline methods with a 4 % improvement in average classification accuracy. Additionally, we identified site-specific brain disease-related biomarkers, offering novel insights into early diagnosis. Code is available at <span><span>https://github.com/ChaojunZ/TDPFL.git</span><svg><path></path></svg></span></div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107987"},"PeriodicalIF":6.3,"publicationDate":"2025-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144889119","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 feature discrimination with pseudo-labels for foundation model in segmentation of 3D medical images","authors":"Ge Jin ,&nbsp;Qian Zhang ,&nbsp;Yong Cheng ,&nbsp;Ming Xu ,&nbsp;Yingwen Zhu ,&nbsp;De Yu ,&nbsp;Yongqi Yuan ,&nbsp;Juncheng Li ,&nbsp;Jun Shi","doi":"10.1016/j.neunet.2025.107979","DOIUrl":"10.1016/j.neunet.2025.107979","url":null,"abstract":"<div><div>Development of medical image segmentation foundation models relies on large-scale samples. However, it is more time-consuming to annotate 3D medical images than 2D natural images, making it challenging to collect sufficient annotated samples. While pseudo-labeling offers a potential solution to expand the annotated dataset, it may introduce noisy labels that can create systematic biases, particularly affecting the segmentation performance of smaller anatomical structures. To this end, we propose a pseudo-label enriched segmentation framework (PESF), which integrates confidence filtering and perturbation-based curriculum learning. To begin with, our pseudo-labeling approach applies a well-pretrained foundation model to generate pseudo-labels for previously unannotated organ categories, effectively expanding the number of classes in the original dataset. Subsequently, we develop a confidence-based filtering mechanism, leveraging a feature extraction module combined with a confidence prediction module to quantitatively assess and filter out low-quality pseudo-labels, thereby minimizing the detrimental effects of noisy pseudo-labels on the model’s optimization. Furthermore, a progressive sampling strategy that integrates curriculum learning with Gaussian random perturbations is proposed, systematically introducing training samples from simpler to more complex cases, thereby enhancing the model’s generalization capability across organs of varying shapes and sizes. Additionally, our theoretical analysis reveals that incorporating these extra pseudo-labeled classes strengthens feature discrimination by increasing the angular margins between class decision boundaries in the embedding space. Experimental results demonstrate that PESF achieves a 6.8% improvement in the overall average Dice Similarity Coefficient (DSC) compared to the baseline SAM-Med3D on (Amos, FLARE22, WORD, BTCV), with particularly gains in challenging anatomical structures such as the pancreas and esophagus. The code is available at <span><span>https://github.com/lonezhizi/PESF</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107979"},"PeriodicalIF":6.3,"publicationDate":"2025-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144864551","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":"Recurrent stochastic configuration networks with block increments","authors":"Dianhui Wang ,&nbsp;Gang Dang","doi":"10.1016/j.neunet.2025.107986","DOIUrl":"10.1016/j.neunet.2025.107986","url":null,"abstract":"<div><div>Recurrent stochastic configuration networks (RSCNs) have shown promise in modelling nonlinear dynamic systems with order uncertainty due to their advantages of easy implementation, less human intervention, and strong approximation capability. This paper develops the original RSCNs with block increments, termed block RSCNs (BRSCNs), to further enhance the learning capacity and efficiency of the network. BRSCNs can simultaneously add multiple reservoir nodes (subreservoirs) during the construction. Each subreservoir is configured with a unique structure in the light of a supervisory mechanism, ensuring the universal approximation property. The reservoir feedback matrix is appropriately scaled to guarantee the echo state property of the network. Furthermore, the output weights are updated online using a projection algorithm, and the persistent excitation conditions that facilitate parameter convergence are also established. Numerical results over a time series prediction, a nonlinear system identification task, and two industrial data predictive analyses demonstrate that the proposed BRSCN performs favourably in terms of modelling efficiency, learning, and generalization performance, highlighting their significant potential for coping with complex dynamics.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107986"},"PeriodicalIF":6.3,"publicationDate":"2025-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144880032","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":"Learning in PINNs: Phase transition, diffusion equilibrium, and generalization","authors":"Sokratis J. Anagnostopoulos ,&nbsp;Juan Diego Toscano ,&nbsp;Nikolaos Stergiopulos ,&nbsp;George Em Karniadakis","doi":"10.1016/j.neunet.2025.107983","DOIUrl":"10.1016/j.neunet.2025.107983","url":null,"abstract":"<div><div>We investigate the learning dynamics of fully-connected neural networks through the lens of the neural gradient signal-to-noise ratio (SNR), examining the behavior of first-order optimizers in non-convex objectives. By interpreting the drift/diffusion phases as proposed in the information bottleneck theory, we identify a third phase termed “diffusion equilibrium” (DE), a stable training phase characterized by highly-ordered neural gradients across the sample space. This phase is marked by an abrupt (first-order) transition, where sample-wise gradients align (SNR increases), and stable optimizer convergence. Moreover, we find that when homogeneous residuals are also met across the sample space during the DE phase, this leads to better generalization, as the optimization steps are equally sensitive to each sample. Based on this observation, we propose a sample-wise re-weighting scheme, which considerably improves the residual homogeneity and generalization in quadratic loss functions, by targeting the problematic samples with large residuals and vanishing gradients. Finally, we explore the information compression phenomenon, pinpointing a significant saturation-induced compression of activations at the DE phase transition, driven by the sample-wise gradient directional alignment. Interestingly, it is during the saturation of activations that the model converges, with deeper layers experiencing negligible information loss. Supported by experimental examples on physics-informed neural networks (PINNs), which highlight the critical role of gradient agreement due to their inherent PDE-based interdependence of samples, our findings suggest that when both sample-wise gradients and residuals transition in an ordered state, this leads to faster convergence and better generalization. Identifying these phase transitions could improve deep learning optimization strategies, enhancing physics-informed methods and overall machine learning performance.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107983"},"PeriodicalIF":6.3,"publicationDate":"2025-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144913018","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":"DRG: A dual relational graph framework for course recommendation","authors":"Yong Ouyang ,&nbsp;Zhen Ye ,&nbsp;Lingyu Chen ,&nbsp;Huanwen Wang ,&nbsp;Yawen Zeng","doi":"10.1016/j.neunet.2025.107984","DOIUrl":"10.1016/j.neunet.2025.107984","url":null,"abstract":"<div><div>The course recommendation system is a core application of recommendation technology in the educational field. Its significance lies in accurately matching users’ interests and needs while providing valuable feedback to instructors, thereby fostering continuous improvement in teaching quality. Various techniques have been proposed for this purpose, with Large Language Models (LLMs) demonstrating significant potential in course recommendation tasks. However, the issue of data sparsity remains a critical bottleneck that limits the accuracy of the recommendation. In this study, we propose a Dual Relationship Graph (DRG) framework that addresses data sparsity by modeling both course-course and user-course relationships through a dual-graph structure. Specifically, DRG constructs two relational graphs: a course-based graph built using LLM-based semantic reasoning, collaborative filtering, clustering, and association rule mining; and a user-based graph constructed via collaborative filtering and LLM-based preference inference. These graphs are integrated into a unified recommendation pipeline through joint graph learning and collaborative reasoning. The enhanced interaction graphs significantly alleviated sparsity, increasing link coverage by 37.88 % and 12.67 % on the two datasets, respectively. Notably, DRG is designed as a plug-and-play module, compatible with both traditional models and LLM-based recommendation systems. Experimental results show that our DRG excels in task ranking across two benchmark datasets, significantly enhancing traditional recommendation models and LLM-based methods. Moreover, DRG’s dual relationship graph consistently outperforms single relationship approaches, underscoring the importance of multi-perspective integration in course recommendation systems. By unifying dual-perspective graph modeling with LLM-driven semantic understanding, DRG provides a scalable and effective solution for personalized course recommendation in sparse educational environments. The code and datasets will be made available at <span><span>https://github.com/WHCK1102/DRG</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107984"},"PeriodicalIF":6.3,"publicationDate":"2025-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144879976","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":"Tensorized anchor alignment for incomplete multi-view clustering","authors":"Yiran Cai ,&nbsp;Hangjun Che ,&nbsp;Wei Guo ,&nbsp;Baicheng Pan ,&nbsp;Man-Fai Leung","doi":"10.1016/j.neunet.2025.107981","DOIUrl":"10.1016/j.neunet.2025.107981","url":null,"abstract":"<div><div>Incomplete Multi-View Clustering (IMVC) focuses on uncovering the consensus and complementary information present in datasets with multiple incomplete views. However, existing IMVC methods face several limitations. First, many approaches exhibit high computational complexity. Second, anchor misalignment across views remains a challenge. Third, high-order correlations among views are often overlooked. To address these challenges, the paper introduces a novel framework called Tensorized Anchor Alignment for Incomplete Multi-view Clustering (TAA-IMC). Specifically, the view-specific anchor graphs are constructed to reduce computational complexity while preserving the diversity of information among views. Then, to mitigate the issue of anchor misalignment, a binary alignment matrix is introduced, ensuring proper correspondence between anchors across different views. Moreover, the aligned anchor graphs are integrated into a tensor representation with a low-rank constraint, enabling the extraction of high-order correlation information. Finally, the proposed TAA-IMC is solved using an alternating update method, showcasing efficiency through memory and time complexity analyses. Extensive comparative experiments conducted on seven benchmark datasets validate the efficiency and superiority of TAA-IMC over state-of-the-art methods.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107981"},"PeriodicalIF":6.3,"publicationDate":"2025-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144885540","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":"SPD-Updater: Symmetric positive definite manifold geometry based temporal updating for visual object tracking","authors":"Jinglin Zhou ,&nbsp;Tianyang Xu ,&nbsp;Xuefeng Zhu ,&nbsp;Xiao-Jun Wu ,&nbsp;Josef Kittler","doi":"10.1016/j.neunet.2025.107985","DOIUrl":"10.1016/j.neunet.2025.107985","url":null,"abstract":"<div><div>Visual object tracking has witnessed continuous advances in recent years along with the exciting developments in backbone networks. In general, all advanced solutions adhere to the template-based tracking framework, which exhibits powerful representative capacity gained via offline training. However, when the target undergoes appearance changes or occlusion, the tracker, which relies on a fixed template defined in the initial frame, struggles to locate it accurately in such complex situations. To achieve online adaptation, recent studies have introduced dynamic templates. Typically, the adopted solution is to compute reliability scores in the traditional Euclidean space to assess the confidence of the dynamic template. However, the Euclidean metric is unreliable to some extent in high-dimensional feature spaces, potentially resulting in a negative impact by involving incorrect dynamic templates. To overcome this problem, we exploit the compact geometric representation capacity of the Symmetric Positive Definite (SPD) manifold to design a novel score prediction module for the tracker update (SPD-Updater). By switching to an SPD manifold metric, we obtain a more accurate and stable dynamic template, thereby enhancing the model capacity to handle complex situations. To validate the reliability of manifold metric in tracking models, we conduct experiments with trackers using different backbones. The experimental results on LaSOT, GOT-10k, TrackingNet, and UAV123 demonstrate the effectiveness of our approach, reflecting the merit of the SPD metric in online tracking adaptation.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"193 ","pages":"Article 107985"},"PeriodicalIF":6.3,"publicationDate":"2025-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144880033","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}