Neural Networks最新文献

{"title":"Sampled-data control-based stabilization of fuzzy inertial quaternion-valued delayed neural networks with parameter uncertainties","authors":"Ziye Zhang ,&nbsp;Shuwen Lv ,&nbsp;Runan Guo ,&nbsp;Zhen Wang ,&nbsp;Chong Lin","doi":"10.1016/j.neunet.2025.107880","DOIUrl":"10.1016/j.neunet.2025.107880","url":null,"abstract":"<div><div>This paper addresses the exponential stabilization of fuzzy inertial quaternion-valued neural networks (FIQVNNs) with time-varying delay and parametric uncertainties. To this end, a non-fragile sampled-data controller incorporating a time-delay term is then designed for the first time to ensure the stability of FIQVNNs. Later, to further reduce conservatism, the improved reciprocally convex inequality is extended to quaternion domain. Furthermore, through constructing appropriate Lyapunov-Krasovskii functionals (LKFs) and utilizing advanced inequality techniques, a refined analytical framework is developed. The derived sufficient stability conditions are presented in the form of linear matrix inequalities (LMIs), which provide standards for system stability analysis. Finally, the proposed results are validated through both numerical simulations and an application example.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"192 ","pages":"Article 107880"},"PeriodicalIF":6.3,"publicationDate":"2025-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144757389","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":"Regional crowd flow estimation from aerial view","authors":"Huibin Wei ,&nbsp;Qi Li ,&nbsp;Xindai Lin ,&nbsp;Yuhao Lin ,&nbsp;Shu Wang ,&nbsp;Shengfeng He ,&nbsp;Antoni B. Chan ,&nbsp;Wenxi Liu","doi":"10.1016/j.neunet.2025.107907","DOIUrl":"10.1016/j.neunet.2025.107907","url":null,"abstract":"<div><div>In recent years, crowd analysis has been widely studied due to its realistic applications in many areas. In this paper, we pose a novel challenge for monitoring large-scale crowd scenes from an aerial view via estimating specific crowd flow for each partitioned region. However, existing methods are difficult to estimate the specific crowd flow for each region flexibly, simply, and accurately, especially lacking clear crowd appearance features from a top-down view. To accomplish this, we present a crowd flow estimation model whose goal is to estimate the flow into and out-of a certain region over any given time span. Specifically, we set up a two-stream network that jointly regresses crowd density and individual velocities, so as to directly approximate the instantaneous flow at each location. To enhance the flow estimation, we utilize the local relationships between crowd distribution and individual velocities via a proposed locality-confined attention module. Furthermore, we incorporate the additional spatio-temporal regularization for the top-down view by reversing future frames via the proposed inverse-temporal loss. In experiments, we apply drone-based overhead crowd videos to evaluate our approach in the task of crowd flow estimation, and we show that our approach surpasses the performance of prior methods and can also be applied in a variety of crowd analysis applications for understanding social scenes.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"192 ","pages":"Article 107907"},"PeriodicalIF":6.3,"publicationDate":"2025-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144749519","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":"MSB-VQA: Overcoming multiple source biases for robust visual question answering","authors":"Jingliang Gu ,&nbsp;Xingjie Zhuang ,&nbsp;Zhixin Li","doi":"10.1016/j.neunet.2025.107908","DOIUrl":"10.1016/j.neunet.2025.107908","url":null,"abstract":"<div><div>Recent studies have found that many VQA models are influenced by biases and cannot effectively utilize multimodal information for reasoning. Models that perform well on standard VQA datasets perform poorly on bias-sensitive VQA-CP datasets. Although there have been many studies focusing on mitigating biases in VQA models, most of them only consider language bias and fail to achieve satisfactory results. To address this issue, we propose a novel method that targets various sources of bias. Specifically, to eliminate multimodal shortcut biases, we design a bias detector, which can be trained by generative adversarial networks and knowledge distillation to use unimodal information to imagine another modality, effectively simulating the process of bias formation in humans. To combat distributional bias, we use a cosine classifier to obtain a cosine feature branch from the base model. We then use adaptive angular margin loss and supervised contrastive loss to address bias caused by uneven sample distributions in terms of frequency, difficulty, and answer. During the prediction phase, we fuse the predictions of the cosine classifier with those of the base model, balancing the model’s performance on ID and OOD datasets. Finally, we conduct extensive experiments on the VQA-CPv2, VQAv2, and VQA-CE datasets, demonstrating that our MSB-VQA method outperforms other methods in bias reduction significantly, without using any data balancing and augmentation.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"192 ","pages":"Article 107908"},"PeriodicalIF":6.3,"publicationDate":"2025-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144757470","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":"Quasi-synchronization of Caputo-Hadamard fractional-order memristive neural networks with time-varying delays","authors":"Haining Li ,&nbsp;Hong-Li Li ,&nbsp;Tingwen Huang ,&nbsp;Xinzhi Liu ,&nbsp;Jinde Cao","doi":"10.1016/j.neunet.2025.107900","DOIUrl":"10.1016/j.neunet.2025.107900","url":null,"abstract":"<div><div>In this paper, quasi-synchronization issue of Caputo-Hadamard fractional memristive neural networks with time-varying delays is investigated. Firstly, on the basis of the definitions and properties of Caputo-Hadamard fractional-order derivative and Mittag-Leffler function, a novel inequality is established, which serves to explore synchronization issues for Caputo-Hadamard fractional-order dynamical networks. Then, under a discontinuous control strategy, quasi-synchronization criteria are obtained by using our constructed inequality and some analytical techniques. Eventually, the effectiveness of the developed theoretical result is validated through a numerical example.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"192 ","pages":"Article 107900"},"PeriodicalIF":6.3,"publicationDate":"2025-07-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144757388","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":"Optimizing the multi-objective traveling salesman problem with a deep reinforcement learning algorithm using cross fusion attention networks","authors":"Xiaoyu Fu ,&nbsp;Shenshen Gu ,&nbsp;Chee-Meng Chew","doi":"10.1016/j.neunet.2025.107904","DOIUrl":"10.1016/j.neunet.2025.107904","url":null,"abstract":"<div><div>The multi-objective traveling salesman problem (MOTSP), a classical type of multi-objective combinatorial optimization problem (MOCOP), is pivotal in numerous real-world applications. However, traditional algorithms often face challenges in efficiently finding satisfactory solutions due to the vast search space and inherent conflicts between objectives. To address this issue, we propose a deep reinforcement learning (DRL) algorithm utilizing a cross fusion attention network (CFAN). The cross fusion attention encoder within the CFAN architecture is designed to capture the relationships between problem instances and weight preferences, thereby constructing unified context features. This enables a single trained CFAN model to solve problems with varying weight preferences. Furthermore, we enhance the model’s ability to explore boundary solutions by adjusting the weight distribution. To evaluate the proposed algorithm’s effectiveness, we conducted a comparative analysis with classical evolutionary algorithms and advanced DRL approaches across various MOTSP instances. Experimental results demonstrate that CFAN consistently outperforms both categories of algorithms, achieving superior solution quality and generalization capability. In particular, CFAN achieves a <span><math><mrow><mn>1.43</mn><mspace></mspace><mo>%</mo></mrow></math></span> improvement in the hypervolume (HV) metric over the best-performing DRL algorithm on KroAB instances, a <span><math><mrow><mn>3.12</mn><mspace></mspace><mo>%</mo></mrow></math></span> improvement on tri-objective problem instances, and a <span><math><mrow><mn>2.17</mn><mspace></mspace><mo>%</mo></mrow></math></span> improvement on large-scale problem instances. These results highlight the effectiveness of CFAN in handling diverse problem instances.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"192 ","pages":"Article 107904"},"PeriodicalIF":6.3,"publicationDate":"2025-07-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144738684","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":"FairForensics: mitigating attribute bias in deepfake detection by integrating texture and attribute features","authors":"Chunlei Peng ,&nbsp;Yinyin Chen ,&nbsp;Decheng Liu ,&nbsp;Nannan Wang ,&nbsp;Xinbo Gao","doi":"10.1016/j.neunet.2025.107899","DOIUrl":"10.1016/j.neunet.2025.107899","url":null,"abstract":"<div><div>With the rapid advancement of artificial intelligence, Deepfake technology, which involves the synthesis of highly realistic face-swapping images and videos, has garnered significant attention. While this technology has various legitimate applications, its misuse in political manipulation, identity fraud, and misinformation poses serious societal risks. Consequently, effective face forgery detection methods are crucial. However, current detection techniques often overlook fairness concerns, with significant disparities observed across different attributes, such as gender and race. These biases not only undermine the reliability of detection systems but also hinder their applicability in diverse cultural contexts. In this paper, we present a novel face forgery detection method named FairForensics, which extracts attribute and texture features to mitigate such biases while enhancing detection accuracy. The method uses the fairtexture module to extract detailed information about facial textures, such as facial skin, hair color, style and wrinkles, which are indicative of forgeries. In parallel, the fairattribute module is introduced to extract high-level semantic features related to gender, race, and other facial attributes. By comparing facial attributes across different video frames, our method identifies inconsistencies that are typical in forged videos, where the facial features may not align consistently over time. On this basis, we design a spatial-temporal feature fair aggregator that effectively integrates texture and attribute features. The interaction between spatial and temporal attention mechanisms captures long-term dependencies, providing a fair feature representation. The proposed approach not only improves detection accuracy but also reduces the detection disparity across different face attributes, effectively mitigating attribute bias. This work contributes to the development of more accurate and fair face forgery detection systems, offering a promising solution to the societal challenges posed by Deepfake technology.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"192 ","pages":"Article 107899"},"PeriodicalIF":6.3,"publicationDate":"2025-07-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144738881","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":"Temporal structure-preserving transformer for industrial load forecasting","authors":"Senzhen Wu ,&nbsp;Zhijin Wang ,&nbsp;Xiufeng Liu ,&nbsp;Yuan Zhao ,&nbsp;Yue Hu ,&nbsp;Yaohui Huang","doi":"10.1016/j.neunet.2025.107887","DOIUrl":"10.1016/j.neunet.2025.107887","url":null,"abstract":"<div><div>Accurate power load forecasting in industrial parks is crucial for optimizing energy management and operational efficiency. Existing models struggle with industrial load series’ complex, multi-target nature and the need to integrate diverse exogenous variables. This paper introduces the Temporal Structure-Preserving Transformer (TSPT), a novel architecture that addresses these challenges by decomposing multi-target series into univariate series, enabling parallel processing and integrating exogenous data. The TSPT model incorporates the Gated Feature Fusion (GFF), which learns to capture multiscale temporal patterns from each target sequence and exogenous factors by preserving the temporal structure of the series. This parallel processing and the structure-preserving transformations allow TSPT to effectively integrate domain-specific knowledge, such as weather, production, and efficiency data, enhancing its forecasting performance. Comprehensive experiments on a real-world industrial park dataset demonstrate TSPT’s superiority over state-of-the-art methods in handling complex, multi-target forecasting tasks with integrated exogenous variables. The proposed approach offers a pathway for scalable and accurate load forecasting in industrial settings, improving energy management and operational decision-making.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"192 ","pages":"Article 107887"},"PeriodicalIF":6.0,"publicationDate":"2025-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144704467","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 intelligent transportation systems with a more efficient model for long-term traffic predictions based on an attention mechanism and a residual temporal convolutional network","authors":"Selim Reza ,&nbsp;Marta Campos Ferreira ,&nbsp;J.J.M. Machado ,&nbsp;João Manuel R.S. Tavares","doi":"10.1016/j.neunet.2025.107897","DOIUrl":"10.1016/j.neunet.2025.107897","url":null,"abstract":"<div><div>Accurate traffic state prediction is fundamental to Intelligent Transportation Systems, playing a critical role in optimising traffic management, improving mobility, and enhancing the efficiency of transportation networks. Traditional methods often rely on feature engineering, statistical time-series approaches, and non-parametric techniques to model the inherent complexities of traffic states, incorporating external factors such as weather conditions and accidents to refine predictions. However, the effectiveness of long-term traffic state prediction hinges on capturing spatial-temporal dependencies over extended periods. Current models face challenges in dealing with (i) high-dimensional traffic features, (ii) error accumulation for multi-step prediction, and (iii) robustness to external factors effectively. To address these challenges, this study proposes a novel model with a Dynamic Feature Embedding layer designed to transform complex data sequences into meaningful representations and a Deep Linear Projection network that refines these representations through non-linear transformations and gating mechanisms. These two features make the model more scalable when dealing with high-dimensional traffic features. The model also includes a Spatial-Temporal Positional Encoding layer to capture spatial-temporal relationships, masked multi-head attention-based encoder blocks, and a Residual Temporal Convolutional Network to process features and extract short- and long-term temporal patterns. Finally, a Time-Distributed Fully Connected Layer produces accurate traffic state predictions up to 24 timesteps into the future. The proposed architecture uses a direct strategy for multi-step modelling to help predict timesteps non-autoregressively and thus circumvents the error accumulation problem. The model was evaluated against state-of-the-art baselines using two benchmark datasets. Experimental results demonstrated the model’s superiority, achieving up to <span><math><mrow><mn>21.17</mn><mspace></mspace><mo>%</mo></mrow></math></span> and <span><math><mrow><mn>29.30</mn><mspace></mspace><mo>%</mo></mrow></math></span> average improvements in Root Mean Squared Error and <span><math><mrow><mn>3.56</mn><mspace></mspace><mo>%</mo></mrow></math></span> and <span><math><mrow><mn>32.80</mn><mspace></mspace><mo>%</mo></mrow></math></span> improvements in Mean Absolute Error compared to the baselines, respectively. The Friedman Chi-Square statistical test further confirmed the significant performance difference between the proposed model and its counterparts. The adversarial perturbations and random sensor dropout tests demonstrated its good robustness. On top of that, it demonstrated good generalizability through extensive experiments. The model effectively mitigates error accumulation in multi-step predictions while maintaining computational efficiency, making it a promising solution for enhancing Intelligent Transportation Systems.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"192 ","pages":"Article 107897"},"PeriodicalIF":6.3,"publicationDate":"2025-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144738682","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":"Wind power generation forecasting system based on multi-model intelligent fusion strategy and probabilistic forecasting technology","authors":"Yamei Chen ,&nbsp;Jianzhou Wang ,&nbsp;Runze Li ,&nbsp;Bo Zeng ,&nbsp;Haiyan Lu","doi":"10.1016/j.neunet.2025.107884","DOIUrl":"10.1016/j.neunet.2025.107884","url":null,"abstract":"<div><div>Due to the excessive consumption of fossil energy, mankind is facing a series of severe challenges such as resource depletion and environmental deterioration, which makes the development, utilization and promotion of clean energy become an inevitable trend in the world. Wind energy as a typical renewable energy, with its clean, environmental protection characteristics in the field of new energy rapid development. However, due to the intermittent and instantaneous fluctuations of wind power, large-scale wind power grid integration and stable operation of power systems face difficult tests. Therefore, accurate and efficient wind prediction is very important for the stability control and integrated scheduling of wind turbines. Based on data from different wind turbines at the Penmanshiel wind farm on the east coast of Scotland, this paper makes deterministic predictions and uncertainty analyses for the next 24, 48 and 72 hours and proposes an integrated wind power system. In data preprocessing phase, the adaptive decomposition reconstruction strategy combined with fuzzy theory, effectively reduce the noise and fluctuations on the result of the experiment data. On this basis, the optimization algorithm is integrated to carry out parameter fine-tuning and structure optimization. Finally, with the aid of quantile regression and kernel density estimation, a scientific, accurate and stable forecasting system is constructed. Compared with the traditional single model forecast, the system not only quantifies the uncertainty of wind forecast, but also improves the forecast accuracy.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"192 ","pages":"Article 107884"},"PeriodicalIF":6.3,"publicationDate":"2025-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144757477","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 adversarial transferability via transformation inference","authors":"Jiaxin Hu ,&nbsp;Jie Lin ,&nbsp;Xiangyuan Yang ,&nbsp;Hanlin Zhang ,&nbsp;Peng Zhao","doi":"10.1016/j.neunet.2025.107896","DOIUrl":"10.1016/j.neunet.2025.107896","url":null,"abstract":"<div><div>The transferability of adversarial examples has become a crucial issue in black-box attacks. Input transformation techniques have shown considerable promise in enhancing transferability, but existing methods are often limited by their empirical nature, neglecting the wide spectrum of potential transformations. This may limit the transferability of adversarial examples. To address this issue, we propose a novel transformation variational inference attack(TVIA) to improve the diversity of transformations, which leverages variational inference (VI) to explore a broader set of input transformations, thus enriching the diversity of adversarial examples and enhancing their transferability across models. Unlike traditional empirical approaches, our method employs the variational inference of a Variational Autoencoder (VAE) model to explore potential transformations in the latent space, significantly expanding the range of image variations. We further enhance diversity by modifying the VAE’s sampling process, enabling the generation of more diverse adversarial examples. To stabilize the gradient direction during the attack process, we fuse transformed images with the original image and apply random noise. The experiment results on Cifar10, Cifar100, ImageNet datasets show that the average attack success rates (ASRs) of the adversarial examples generated by our TVIA surpass all existing attack methods. Specially, the ASR reaches 95.80 % when transferred from Inc-v3 to Inc-v4, demonstrating that our TVIA can effectively enhance the transferability of adversarial examples.</div></div>","PeriodicalId":49763,"journal":{"name":"Neural Networks","volume":"192 ","pages":"Article 107896"},"PeriodicalIF":6.3,"publicationDate":"2025-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144721597","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}