Neural NetworksPub Date : 2025-04-10DOI: 10.1016/j.neunet.2025.107443
Chen Li , Guoyan Huang , Zhu Sun , Lu Zhang , Shanshan Feng , Guanfeng Liu
{"title":"PCDe: A personalized conversational debiasing framework for next POI recommendation with uncertain check-ins","authors":"Chen Li , Guoyan Huang , Zhu Sun , Lu Zhang , Shanshan Feng , Guanfeng Liu","doi":"10.1016/j.neunet.2025.107443","DOIUrl":"10.1016/j.neunet.2025.107443","url":null,"abstract":"<div><div>In the next point-of-interest (POI) recommendation, users may visit <em>individual POIs</em> within larger gathering places, such as shopping malls (termed as <em>collective POIs</em>), leading to uncertain check-ins. Our data analysis unveils that (1) the presence of such uncertain check-ins raises a new type of bias, termed as <em>scale bias</em>, that is, the recommender tends to recommend collective POIs over individual POIs, which further exacerbates the commonly-observed <em>popularity bias</em>, that is, the recommender tends to recommend popular POIs rather than unpopular ones; and (2) the existence of the above two types of biases significantly affects the fairness of next POI recommendation with uncertain check-ins. Therefore, we propose a <u>P</u>ersonalized <u>C</u>onversational <u>De</u>biasing framework (PCDe) by exploiting the advantages of conversational techniques to capture personalized dynamic user preferences, thereby mitigating both scale and popularity biases at a personalized level. Specifically, the <em>inquiry component</em> designs an improved question-and-answer manner based on personalized information entropy, thus mitigating the scale bias. The <em>rewarding component</em> then introduces a novel debiasing reward mechanism based on the Jensen–Shannon divergence to make the recommendations better aligned with users’ historical preferences on popularity, thereby addressing the popularity bias. Extensive experiments demonstrate the superiority of our proposed PCDe over state-of-the-arts (SOTAs) regarding mitigating scale and popularity biases while enhancing recommendation accuracy thanks to its personalized debiasing mechanism.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107443"},"PeriodicalIF":6.0,"publicationDate":"2025-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143844167","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}
Neural NetworksPub Date : 2025-04-10DOI: 10.1016/j.neunet.2025.107446
Yaoxin Wu , Zhiguang Cao , Wen Song , Yingqian Zhang
{"title":"Solving two-stage stochastic integer programs via representation learning","authors":"Yaoxin Wu , Zhiguang Cao , Wen Song , Yingqian Zhang","doi":"10.1016/j.neunet.2025.107446","DOIUrl":"10.1016/j.neunet.2025.107446","url":null,"abstract":"<div><div>Solving stochastic integer programs (SIPs) is extremely intractable due to the high computational complexity. To solve two-stage SIPs efficiently, we propose a conditional variational autoencoder (CVAE) for scenario representation learning. A graph convolutional network (GCN) based VAE embeds scenarios into a low-dimensional latent space, conditioned on the deterministic context of each instance. With the latent representations of stochastic scenarios, we perform two auxiliary tasks: <em>objective prediction</em> and <em>scenario contrast</em>, which predict scenario objective values and the similarities between them, respectively. These tasks further integrate objective information into the representations through gradient backpropagation. Experiments show that the learned scenario representations can help reduce scenarios in SIPs, facilitating high-quality solutions in a short computational time. This superiority generalizes well to instances of larger sizes, more scenarios, and various distributions.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107446"},"PeriodicalIF":6.0,"publicationDate":"2025-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143850864","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}
Neural NetworksPub Date : 2025-04-09DOI: 10.1016/j.neunet.2025.107457
Dongrui Gao , Mengwen Liu , Haokai Zhang , Manqing Wang , Hongli Chang , Gaoxiang Ouyang , Shihong Liu , Pengrui Li
{"title":"A multi-domain constraint learning system inspired by adaptive cognitive graphs for emotion recognition","authors":"Dongrui Gao , Mengwen Liu , Haokai Zhang , Manqing Wang , Hongli Chang , Gaoxiang Ouyang , Shihong Liu , Pengrui Li","doi":"10.1016/j.neunet.2025.107457","DOIUrl":"10.1016/j.neunet.2025.107457","url":null,"abstract":"<div><div>Neuroscience shows that the brain stimulated by external information can induce functional responses to emotions, which can be measured and analyzed by electroencephalogram (EEG). Most existing works focus on extracting specific spatial topological information and temporal dependency representations, with a few works begin to mine the value of spatiotemporal cross-domain information. However, these approaches overdependence on cognitive prior information, limiting their ability to grasp complex domain-structured information. Moreover, the stable extraction of cognitive functions is crucial for reinforcing the performance of emotion recognition systems. Here, we propose a multi-domain constraint learning system. It is inspired by adaptive cognitive graphs, embedding spatiotemporal representative knowledge into the constructed framework (AC-DCL) to improve the performance of emotion recognition. In the AC-DCL, a spatial-guided dynamic graph constraint learning module is meticulously designed to overcome reliance on cognitive priors and adaptively generate and constrain functional relationships within cognitive graphs. At the same time, a temporal-driven sequence transformer is proposed to extract global temporal dependency features. Furthermore, this study designs a novel multi-domain interactive attention module with constraining domain-specific differences and aggregating complementary information, which surpasses traditional static cross-domain interactions. The essence of the proposed AC-DCL lies in capturing stable cognitive functions from complex and dynamic cognitive structures. Experimental results on the DREAMER, FACED, and SEED-IV datasets demonstrate the impressive advantages of AC-DCL and its potential to drive the learning of cross-domain interaction representations.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107457"},"PeriodicalIF":6.0,"publicationDate":"2025-04-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143817119","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}
Neural NetworksPub Date : 2025-04-09DOI: 10.1016/j.neunet.2025.107456
Huihuang Zhang, Haigen Hu, Deming Zhou, Xiaoqin Zhang, Bin Cao
{"title":"Compact CNN module balancing between feature diversity and redundancy","authors":"Huihuang Zhang, Haigen Hu, Deming Zhou, Xiaoqin Zhang, Bin Cao","doi":"10.1016/j.neunet.2025.107456","DOIUrl":"10.1016/j.neunet.2025.107456","url":null,"abstract":"<div><div>Feature diversity and redundancy play a crucial role in enhancing a model’s performance, although their effect on network design remains underexplored. Herein, we introduce BDRConv, a compact convolutional neural network (CNN) module that establishes a balance between feature diversity and redundancy to generate and retain features with moderate redundancy and high diversity while reducing computational costs. Specifically, input features are divided into a main part and an expansion part. The main part extracts intrinsic and diverse features, while the expansion part enhances diverse information extraction. Experiments on the CIFAR10, ImageNet, and MS COCO datasets demonstrate that BDRConv-equipped networks outperform state-of-the-art methods in accuracy, with significantly lower floating-point operations (FLOPs) and parameters. In addition, BDRConv module as a plug-and-play component can easily replace existing convolution modules, offering potential for broader CNN applications.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107456"},"PeriodicalIF":6.0,"publicationDate":"2025-04-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143817121","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}
Neural NetworksPub Date : 2025-04-09DOI: 10.1016/j.neunet.2025.107452
Rui Meng , Changchun Hua , Kuo Li , Qidong Li
{"title":"Dynamic events-based adaptive NN output feedback control of interconnected nonlinear systems under general output constraint","authors":"Rui Meng , Changchun Hua , Kuo Li , Qidong Li","doi":"10.1016/j.neunet.2025.107452","DOIUrl":"10.1016/j.neunet.2025.107452","url":null,"abstract":"<div><div>This paper investigates the adaptive NN output feedback tracking control problem for a class of interconnected nonlinear systems. Unlike the existing control algorithms, we propose a dynamic event-triggered output constraint control algorithm.First, a reduced-order dynamic gain K-filter is established to construct the unmeasurable state variables. Second, an asymmetric constraint function with a special time-varying function is proposed, which can handle the case where the initial values of the constraint boundaries are unlimited. Then, a dynamic event-triggered mechanism based on the arctangent function is developed, which avoids the continuous transmission of control signals. With the help of the Lyapunov stability theory, it is rigorously proved that all signals of the closed-loop systems are bounded and the tracking error satisfies the output constraint requirement.Finally, the validity of the proposed algorithm is justified by the use of a numerical simulation.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107452"},"PeriodicalIF":6.0,"publicationDate":"2025-04-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143829319","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":"TCH: A novel multi-view dimensionality reduction method based on triple contrastive heads","authors":"Hongjie Zhang , Ruojin Zhou , Siyu Zhao , Ling Jing , Yingyi Chen","doi":"10.1016/j.neunet.2025.107459","DOIUrl":"10.1016/j.neunet.2025.107459","url":null,"abstract":"<div><div>Multi-view dimensionality reduction (MvDR) is a potent approach for addressing the high-dimensional challenges in multi-view data. Recently, contrastive learning (CL) has gained considerable attention due to its superior performance. However, most CL-based methods focus on promoting consistency between any two cross views from the perspective of subspace samples, which extract features containing redundant information and fail to capture view-specific discriminative information. In this study, we propose feature- and recovery-level contrastive losses to eliminate redundant information and capture view-specific discriminative information, respectively. Based on this, we construct a novel MvDR method based on triple contrastive heads (TCH). This method combines sample-, feature-, and recovery-level contrastive losses to extract sufficient yet minimal subspace discriminative information in accordance with the information bottleneck principle. Furthermore, the relationship between TCH and mutual information is revealed, which provides the theoretical support for the outstanding performance of our method. Our experiments on five real-world datasets show that the proposed method outperforms existing methods.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107459"},"PeriodicalIF":6.0,"publicationDate":"2025-04-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143838014","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}
Neural NetworksPub Date : 2025-04-09DOI: 10.1016/j.neunet.2025.107458
Xiao Li , Hongmei Wang , Yitian Xu
{"title":"Safe and accelerated screening framework for support tensor machines","authors":"Xiao Li , Hongmei Wang , Yitian Xu","doi":"10.1016/j.neunet.2025.107458","DOIUrl":"10.1016/j.neunet.2025.107458","url":null,"abstract":"<div><div>Support Tensor Machines (STMs) constitute an effective supervised learning method for classifying high-dimensional tensor data. However, traditional iterative solving methods are often time-consuming. To effectively address the issue of lengthy training times, inspired by the safe screening strategies employed in support vector machines, we generalize the safe screening rule to the tensor domain and propose a novel safe screening rule for STM, which includes the dual static screening rule (DSSR), the dynamic screening rule (DGSR), and a subsequent checking verification. The screening rule initially employs variational inequalities to screen out a portion of redundant samples before training, reducing the problem scale. During the training process, the rule further accelerates training by iteratively screening redundant samples using the duality gap. We also design a subsequent checking technique based on optimality conditions to guarantee the safety of the screening rule. Building on this, we also develop a flexible safe screening framework, referred to as DS-DGSR, which incorporates the DSSR and the DGSR. It not only tackles the challenges of combining various tensor decomposition methods and the diverse scenarios of the decomposed coefficient parameter and decomposed samples in STMs, but also offers flexible adaptation and application according to the characteristics of different STMs. Numerical experiments on multiple real-world high-dimensional tensor datasets confirm the effectiveness and feasibility of DS-DGSR.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107458"},"PeriodicalIF":6.0,"publicationDate":"2025-04-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143833249","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}
Neural NetworksPub Date : 2025-04-09DOI: 10.1016/j.neunet.2025.107430
Qiuye Wu , Bo Zhao , Derong Liu
{"title":"Event-triggered control for input-constrained nonzero-sum games through particle swarm optimized neural networks","authors":"Qiuye Wu , Bo Zhao , Derong Liu","doi":"10.1016/j.neunet.2025.107430","DOIUrl":"10.1016/j.neunet.2025.107430","url":null,"abstract":"<div><div>To accommodate the increasing system scale, improve the system operation success rate and save the computational and communication resources, it is urgent to obtain the Nash equilibrium solution for systems with increasing controllers in an effective way. In this paper, nonzero-sum game problem of partially unknown nonlinear systems with input constraints is solved via the particle swarm optimized neural network-based integral reinforcement learning. By introducing the integral reinforcement learning technique, the drift dynamics is not required any more. To further improve the success rate of system operation, extended adaptive particle swarm optimization algorithm which shares the individual historical optimal position with the whole population is adopted in tuning neural network weights, rather than sharing only the current particle in the traditional particle swarm optimization algorithm. The control policy for each player is obtained by solving the coupled Hamilton–Jacobi equation with a single critic neural network, which simplifies the control structure and reduces the computational burden. Moreover, by introducing the event-triggering mechanism, the control policies are updated at event-triggering instants only. Thus, the computational and communication burdens are further reduced. The stability of the closed-loop system is guaranteed by implementing the integral reinforcement learning-based event-triggered control policies via the Lyapunov’s direct method. From the comparative simulation results, the developed integral reinforcement learning-based event-triggered control scheme via the extended adaptive particle swarm optimization performs better than those using gradient descent algorithm, nonlinear programming, particle swarm optimization and other popular training algorithms.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107430"},"PeriodicalIF":6.0,"publicationDate":"2025-04-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143807969","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}
Neural NetworksPub Date : 2025-04-08DOI: 10.1016/j.neunet.2025.107478
Wenzhuo Liu , Shuiying Xiang , Tao Zhang , Yanan Han , Yahui Zhang , Xingxing Guo , Licun Yu , Yue Hao
{"title":"S4-KD: A single step spiking SiamFC+ + for object tracking with knowledge distillation","authors":"Wenzhuo Liu , Shuiying Xiang , Tao Zhang , Yanan Han , Yahui Zhang , Xingxing Guo , Licun Yu , Yue Hao","doi":"10.1016/j.neunet.2025.107478","DOIUrl":"10.1016/j.neunet.2025.107478","url":null,"abstract":"<div><div>Spiking neural networks (SNNs), which transmit information through binary spikes, have the advantages of high efficiency and low energy consumption. At present, the multiple time steps of SNNs can lead to increased latency and power consumption. To this end, we propose Single Step Spiking SiamFC+ + (S4), an improved single-step end-to-end direct training target tracking framework that compresses the time step to 1 by temporal pruning, using AlexNet as the backbone network. Experimental results show that, even when only a single time step is used, the tracking performance of the proposed S4 is still comparable to the original Spiking SiamFC+ +. Furthermore, we introduce the knowledge distillation to improve the performance of the proposed S4, which is called S4-KD for clarity. Three kinds of distillation loss functions are designed for the S4-KD. An artificial neural network model based on the AlexNet network serves as the teacher model, while the temporal-pruned S4 model acts as the student model for retraining. Experimental results show that the S4-KD tracker achieves higher performance on several tracking benchmarks. More specifically, on the OTB100 dataset, Precision and Success are 0.871 and 0.657 respectively, on the UAV123 dataset, Precision and Success are 0.766 and 0.603 respectively, and on the VOT2018 dataset, A, R, and EAO are 0.582, 0.370, and 0.278 respectively. In addition, the estimated energy consumption of the S4-KD is only 34.6 % of that of the original Spiking SiamFC+ +. To the best of our knowledge, the proposed S4-KD tracker surpasses all the existing SNN-based object tracking methods, achieving state-of-the-art performance. Our codes will be available at <span><span>https://github.com/PSNN-xd/S4-KD</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107478"},"PeriodicalIF":6.0,"publicationDate":"2025-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143829318","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":"PILOT: Deep Siamese network with hybrid attention improves prediction of mutation impact on protein stability","authors":"Yuan Zhang , Junsheng Deng , Mingyuan Dong , Jiafeng Wu , Qiuye Zhao , Xieping Gao , Dapeng Xiong","doi":"10.1016/j.neunet.2025.107476","DOIUrl":"10.1016/j.neunet.2025.107476","url":null,"abstract":"<div><div>Evaluating the mutation impact on protein stability (ΔΔ<em>G</em>) is essential in the study of protein engineering and understanding molecular mechanisms of disease-associated mutations. Here, we propose a novel deep learning framework, PILOT, for improved prediction of ΔΔ<em>G</em> using a Siamese network with hybrid attention mechanism. The PILOT framework leverages multiple attention modules to effectively extract representations for amino acids, atoms, and protein sequences, respectively. This approach significantly ensures the deep fusion of structural information at both residue and atom levels, the seamless integration of structural and sequence representations, and the effective capture of both long-range and short-range dependencies among amino acids. Our extensive evaluations demonstrate that PILOT greatly outperforms other state-of-the-art methods. We also showcase that PILOT identifies exceptional patterns for different mutation types. Moreover, we illustrate the clinical applicability of PILOT in highlighting pathogenic variants from benign variants and VUS (variants of uncertain significance), and distinguishing de novo mutations in disease cases and controls. In summary, PILOT presents a robust deep learning tool that could offer significant insights into drug design, medical applications, and protein engineering studies.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"188 ","pages":"Article 107476"},"PeriodicalIF":6.0,"publicationDate":"2025-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143844169","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}