Applied Intelligence最新文献

{"title":"A reinforcement learning malware detection model based on heterogeneous information network path representation","authors":"Kang Yang,&nbsp;Lizhi Cai,&nbsp;Jianhua Wu,&nbsp;Zhenyu Liu,&nbsp;Meng Zhang","doi":"10.1007/s10489-025-06417-1","DOIUrl":"10.1007/s10489-025-06417-1","url":null,"abstract":"<div><p>With the significant increase of Android malware, the APP privacy data leakage incidents occur frequently, which poses a great threat to user property and information security. Specifically, the new malware has the characteristics of high evolution rate and diverse variants, leading to the fact that the current malware detection methods still have three key problems: (1) Difficulty in acquiring Android sample structural features; (2) Weakly in representing malware behavior structure; (3) Poor robustness of the detection model. To address the above limitations, we propose a new malware detection framework <b>MPRLDroid</b> with reinforcement learning. First of all, the MPRLDroid model extracts the Android APP structural features and constructs the heterogeneous information network data based on the semantic call structure between APP, API and permission. Subsequently, the model utilizes reinforcement learning to adaptively generate a meta-path for each sample and combines it with a graph attention network to effectively represent the graph of nodes. Finally, the low-dimensional graph node vector data is brought into the downstream detection task for classification, where the performance change of the classification result is used as a reward function for reinforcement learning. The experimental results demonstrate that the MPRLDroid model, when integrated with reinforcement learning, outperforms the baseline models in terms of performance, and its detection model exhibits greater robustness compared to other models.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143632468","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"MSF-SegFormer: a feature fusion algorithm for magnetic leakage image segmentation","authors":"Zhujun Wang,&nbsp;Rongtai Ni,&nbsp;Tianhe Sun,&nbsp;Yulong Jiang,&nbsp;Bin Liu","doi":"10.1007/s10489-025-06453-x","DOIUrl":"10.1007/s10489-025-06453-x","url":null,"abstract":"<div><p>Traditional segmentation networks have low segmentation accuracy for flux leakage images, often leading to missed or false detections of small defects, which significantly affect the evaluation of defect severity. Based on the SegFormer network, a high-accuracy decoder based on multi-scale feature fusion is proposed, which is more suitable for the segmentation of small defects in flux leakage and replaces the multi-layer perceptron (MLP) decoder of the original network. The new network model is called MSF-SegFormer. MSF-SegFormer introduces a feature fusion network MSF that integrates high-resolution and low-resolution features and introduces feature pyramid fusion, which can merge output features at different levels across different scales. A cascaded attention module is proposed, combining two local attention mechanisms in a cascade and using a residual network to enhance the local feature representation of flux leakage images, improving the accuracy and stability of the task. In the application of flux leakage defect data, compared with benchmark models such as CNN and SegFormer, this model can accurately segment target edges with fewer parameters, maintain high accuracy, reduce false detection probability, and improve the Miou value of the traditional MLP decoder from 88.21% to 90.44%.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143632576","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"Performance of machine learning methods for cattle identification and recognition from retinal images","authors":"Pınar Cihan,&nbsp;Ahmet Saygılı,&nbsp;Muhammed Akyüzlü,&nbsp;Nihat Eren Özmen,&nbsp;Celal Şahin Ermutlu,&nbsp;Uğur Aydın,&nbsp;Alican Yılmaz,&nbsp;Özgür Aksoy","doi":"10.1007/s10489-025-06398-1","DOIUrl":"10.1007/s10489-025-06398-1","url":null,"abstract":"<div><p>Animal identification is a critical issue in terms of security, traceability, and animal health, especially in large-scale livestock enterprises. Traditional methods (such as ear tags and branding) both negatively affect animal welfare and may lead to security vulnerabilities. This study aims to develop a biometric system based on retinal vascular patterns for the identification and recognition of cattle. This system aims to provide a safer and animal welfare-friendly alternative by using image processing techniques instead of traditional device-based methods. In the study, preprocessing, segmentation, feature extraction, and performance evaluation steps were applied for the biometric identification and recognition process using retinal images taken from both eyes. Techniques such as green channel extraction, contrast-limited adaptive histogram equalization, morphological operations, noise filtering, and threshold determination were used in the preprocessing stage. Fuzzy C-means, K-means, and Level-set methods were applied for segmentation, and feature extraction was performed using SIFT, SURF, BRISK, FAST, and HARRIS methods. At the end of the study, the highest accuracy rate was obtained as 95.6% for identification and 87.9% for recognition. In addition, the obtained dataset was shared publicly, thus creating a reusable resource that researchers from different disciplines can use. It was concluded that this study made a significant contribution to the field of biometric-based animal identification and recognition and offered a practically usable solution in terms of animal welfare and safety.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10489-025-06398-1.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143622243","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"SFP: temporal knowledge graph completion based on sequence-focus patterns representation learning","authors":"Jingbin Wang,&nbsp;XiFan Ke,&nbsp;FuYuan Zhang,&nbsp;YuWei Wu,&nbsp;SiRui Zhang,&nbsp;Kun Guo","doi":"10.1007/s10489-025-06306-7","DOIUrl":"10.1007/s10489-025-06306-7","url":null,"abstract":"<div><p>The extrapolation task in the temporal knowledge graph has received increasing attention from scholars due to its wide range of practical application scenarios. At present, recurrent neural networks are currently widely used in temporal knowledge graph completion techniques. These networks are employed to depict the sequential pattern of entities and relations. However, as the sequence lengthens, some critical early information may become diluted. Prediction errors ensue in the completion task as a result. Furthermore, it is observed that existing temporal knowledge graph completion methods fail to account for the topological structure of relations, which leads to relation representations with essentially little distinction across different timestamps. In order to tackle the previously mentioned concern, our research introduces a Temporal Knowledge Graph Completion Method utilizing Sequence-Focus Patterns Representation Learning (SFP). This method contains two patterns: the Focus pattern and the Sequential pattern. In the SFP model, we developed a novel graph attention network called ConvGAT. This network efficiently distinguishes and extracts complex relation information, thereby enhancing the accuracy of entity representations that are aggregated in the Focus pattern and Sequential pattern. Furthermore we proposed RelGAT, a graph attention network that simulates the topological structure of relations. This enhances the precision of relation representations and facilitates the differentiation between relation embeddings generated at various timestamps in the Focus pattern. Utilizing a time-aware attention mechanism, the Focus pattern extracts vital information at particular timestamps in order to amplify the data that the Sequential pattern dilutes. On five distinct benchmark datasets, SFP significantly outperforms the baseline, according to a comprehensive series of experiments.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143622099","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"Uncertainty weighted policy optimization based on Bayesian approximation","authors":"Tianyi Li,&nbsp;Genke Yang,&nbsp;Jian Chu","doi":"10.1007/s10489-025-06303-w","DOIUrl":"10.1007/s10489-025-06303-w","url":null,"abstract":"<div><p>Efficient exploration remains a major challenge in the field of reinforcement learning (RL). Bayesian methods have been widely investigated within the RL paradigm and are used to implement intelligent exploration strategies. However, most of these methods inevitably introduce some complexity within the Bayesian neural networks (BNNs) or are difficult to optimize elegantly. In this work, a novel algorithm called uncertainty weighted policy optimization (UWPO) based on Bayesian approximation, is introduced. UWPO theoretically analyzes the uncertainty of the policy space using the Dirichlet distribution and Monte Carlo (MC) dropout for both discrete and continuous spaces, eliminating the need for an explicit distribution representation in BNNs. The algorithm also proposes an implicit distributional training method for the value function, which is compatible with Bayesian inference. Moreover, an uncertainty-weighted update principle is adopted to adaptively adjust the contribution of each training instance to the objective. Finally, comparing UWPO with other prevailing deep reinforcement learning (DRL) algorithms on the Atari, MuJoCo, and Box2D platforms. The experimental results demonstrate that the algorithm improves the average reward score by nearly 15% while reducing computational costs by 20% compared to current state-of-the-art methods.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143622245","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"QuinNet: Quintuple u-shape networks for scale- and shape-variant lesion segmentation","authors":"Gaojuan Fan,&nbsp;Jie Wang,&nbsp;Ruixue Xia,&nbsp;Funa Zhou,&nbsp;Chongsheng Zhang","doi":"10.1007/s10489-025-06448-8","DOIUrl":"10.1007/s10489-025-06448-8","url":null,"abstract":"<div><p>Deep learning approaches have demonstrated remarkable efficacy in medical image segmentation. However, they continue to struggle with challenges such as the loss of global context information, inadequate aggregation of multi-scale context, and insufficient attention to lesion regions characterized by diverse shapes and sizes. To address these challenges, we propose a new medical image segmentation network, which consists of one main U-shape network (MU) and four auxiliary U-shape sub-networks (AU), leading to Quintuple U-shape networks in total, thus abbreviated as <i>QuinNet</i> hereafter. MU devises special attention-based blocks to prioritize important regions in the feature map. It also contains a multi-scale interactive aggregation module to aggregate multi-scale contextual information. To maintain global contextual information, AU encoders extract multi-scale features from the input images, then fuse them into feature maps of the same level in MU, while the decoders of AU refine features for the segmentation task and co-supervise the learning process with MU. Overall, the dual supervision of MU and AU is very beneficial for improving the segmentation performance on lesion regions of diverse shapes and sizes. We validate our method on four benchmark datasets, showing that it achieves significantly better segmentation performance than the competitors. Source codes of QuinNet are available at https://github.com/Truman0o0/QuinNet.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143622232","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"Effect of continuous S-shaped rectified linear function on deep convolutional neural network","authors":"Anahita Ghazvini,&nbsp;Siti Norul Huda Sheikh Abdullah,&nbsp;Masri Ayob","doi":"10.1007/s10489-025-06399-0","DOIUrl":"10.1007/s10489-025-06399-0","url":null,"abstract":"<p>The vanishing gradient issue in convolutional neural networks (CNNs) is often addressed by improving activation functions, such as the S-shaped rectified linear activation unit (SReLU). However, SReLU can pose challenges in updating training parameters effectively. To mitigate this, we propose applying the Aggregation Fischer–Burmeister (AFB) function to SReLU, which smooths the secant line slope of the function from both sides. However, direct application of AFB to SReLU can intensify the vanishing gradient issue due to irregular function behavior. To address this concern, we introduce a regulated version of AFB (ReAFB) that ensures proper gradient and mean activation output conditions when applied to SReLU (ReAFBSReLU). We evaluate the performance of CNNs using ReAFBSReLU on three benchmark datasets: MNIST, CIFAR-10 (with and without data augmentation), and CIFAR-100. Specifically, we implement Network in Network (NIN) for MNIST and CIFAR-10, and LeNet for CIFAR-100 dataset. Additionally, we utilize SqueezeNet exclusively to compare the performance of CNNs using the proposed ReAFBSReLU activation function against state-of-the-art activation functions. Our results demonstrate that ReAFBSReLU outperforms other activation functions tested in this study, indicating its efficacy in enhancing training parameter updates and subsequently improving accuracy.</p>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143622244","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"Use of artificial intelligence techniques in characterization of vibration signals for application in agri-food engineering","authors":"Amalia Luque,&nbsp;Daniel Campos Olivares,&nbsp;Mirko Mazzoleni,&nbsp;Antonio Ferramosca,&nbsp;Fabio Previdi,&nbsp;Alejandro Carrasco","doi":"10.1007/s10489-025-06424-2","DOIUrl":"10.1007/s10489-025-06424-2","url":null,"abstract":"<div><p>Bottling machinery is a critical component in agri-food industries, where maintaining operational efficiency is key to ensuring productivity and minimizing economic losses. Early detection of faulty conditions in this equipment can significantly improve maintenance procedures and overall system performance. This research focuses on health monitoring of gripping pliers in bottling plants, a crucial task that has traditionally relied on analyzing raw vibration signals or using narrowly defined, application-specific features. However, these methods often face challenges related to limited robustness, high computational costs, and sensitivity to noise. To address these limitations, we propose a novel approach based on generic features extracted through basic signal processing techniques applied to vibration signals. These features are then classified using a random forest algorithm, enabling an effective analysis of health states. The proposed method is evaluated against traditional approaches and demonstrates clear advantages, including higher accuracy in detecting and classifying faulty conditions, greater robustness against random perturbations, and a reduced computational cost. Additionally, the method requires fewer training instances to achieve reliable performance. This study highlights the potential of artificial intelligence and signal processing techniques in predictive maintenance, offering a scalable and efficient solution for fault detection in manufacturing processes, particularly within the agri-food sector.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10489-025-06424-2.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143622233","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"AI-computing, deep reinforcement learning-based predictive human-robot neuromechanical simulation for wearable robots","authors":"Mingyi Wang,&nbsp;Shuzhen Luo","doi":"10.1007/s10489-025-06360-1","DOIUrl":"10.1007/s10489-025-06360-1","url":null,"abstract":"<div><p>Human-robot interaction (HRI) is widely used in robotics to assist humans, with wearable robots enhancing mobility for both able-bodied individuals and those with impairments. Traditionally, characterizing human biomechanical responses to these robots requires extensive human testing, which is time-consuming, costly, and potentially risky. Developing computational HRI simulations for wearable robots offers a promising solution. However, modeling the high-fidelity human-exoskeleton interaction in simulations presents significant challenges that remain underexplored. These include creating a high-fidelity autonomous human motion control agent, accounting for the non-passive nature of human responses, and incorporating closed-loop control within the robotic system. In this paper, we propose an AI-computing, deep reinforcement learning-based HRI simulation to predict complex and realistic human biomechanical responses to exoskeleton assistance. The multi-neural network training process develops an end-to-end, autonomous control policy that reduces human muscle effort by utilizing current human kinematic states. This approach processes state information from both the human musculoskeletal and exoskeleton control neural network, generating control policies for robust human walking movement and reducing muscle effort. Numerical experiments demonstrated the framework’s ability to simulate human motion control, showing reductions in hip joint torque (13.04<span>(%)</span>), rectus femoris (RF) muscle activation (7.31<span>(%)</span>), and biceps femoris (BF) muscle activation (12.21<span>(%)</span>) with exoskeleton use. Validation through real-world experiments further confirmed a decrease in RF and BF muscle activations by 22.12<span>(%)</span> and 11.45<span>(%)</span>, respectively. These results highlight the effectiveness of our proposed AI computing-based simulation method in replicating and optimizing human biomechanics during exoskeleton-assisted movement. This AI computing-based human-exoskeleton predictive simulation may offer a general, high-fidelity platform for studying human biomechanical responses and enabling autonomous control for assistive devices without requiring intensive human testing in the rehabilitation field.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143622248","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 comparison study of several strategies in multivariate time series clustering based on graph community detection","authors":"Hanlin Sun,&nbsp;Wei Jie,&nbsp;Yanping Chen,&nbsp;Zhongmin Wang","doi":"10.1007/s10489-025-06444-y","DOIUrl":"10.1007/s10489-025-06444-y","url":null,"abstract":"<div><p>Time series data analysis, especially forecasting, classification, imputation, and anomaly detection, has gained a lot of research attention in recent years due to its prevalence and wide application. Compared to classification, clustering is an unsupervised task and thus more applicable for analyzing massive time series without labels. One latest way is based on the idea of graph community detection: first transforming a time series set into a graph (or a network), in which a node represents a time series instance and an edge denotes that the two connected nodes (thus the represented time series) are more similar to each other; then, running a community detection algorithm on the graph to discover a community structure, that gives out a clustering result. Recently, there are several works based on the graph community detection idea to cluster multivariate time series. However, such works focus only on specific methods in each step, and a performance comparison of combinations of methods in different steps is lacking. This paper outlines the process of graph-based multivariate time clustering as four phases (referred to as framework), namely <i>representation learning</i>, <i>similarity computing</i>, <i>relation network construction</i>, and <i>clustering</i>, lists typical methods in each phase, and makes a comparison study of combinations of each phase methods (called strategies in this paper). Recent time series deep neural network models are introduced to the framework as time series representation learning methods as well. In addition, <span>(varvec{varepsilon } varvec{k})</span>NN, an improvement of <span>(varvec{k})</span>NN by filtering out unnecessary low similarity connections during network construction, is proposed. A great number of experiments are conducted on eight real-world multivariate time series with various properties to verify the performance of different strategy combinations. The results suggest that proper deep neural network is a promising way for learning time series vector representations to compute similarities, and strategies including <span>(varvec{varepsilon } varvec{k})</span>NN for network construction, average for multi-layer network merging and Louvain for clustering are more effective from a statistical perspective.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143612340","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}