检索结果-内蒙古大学图书馆

On-Demand Design of Metasurfaces through Multineural Network Fusion

ACS APPLIED MATERIALS & INTERFACES 2024年第37期16卷 49673-49686页

作者： Li, Junwei Yang, Chengfu Qinhua, A. Lan, Qiusong Yun, Lijun Xia, Yuelong Yunnan Normal Univ Sch Informat Sci & Technol Kunming 650500 Peoples R China Engn Res Ctr Comp Vis & Intelligent Control Techno Dept Educ Yunnan Prov Kunming 650500 Peoples R China

In this paper, a multineural network fusion freestyle metasurface on-demand design method is proposed. The on-demand design method involves rapidly generating corresponding metasurface patterns based on the user-defined spectrum. The generated patterns are then input into a simulator to predict their corresponding S-parameter spectrogram, which is subsequently analyzed against the real S-parameter spectrogram to verify whether the generated metasurface patterns meet the desired requirements. The methodology is based on three neural network models: a Wasserstein Generative Adversarial Network model with a U-net architecture (U-WGAN) for inverse structural design, a variational autoencoder (VAE) model for compression, and an LSTM + Attention model for forward S-parameter spectrum prediction validation. The U-WGAN is utilized for on-demand reverse structural design, aiming to rapidly discover high-fidelity metasurface patterns that meet specific electromagnetic spectrum responses. The VAE, as a probabilistic generation model, serves as a bridge, mapping input data to latent space and transforming it into latent variable data, providing crucial input for a forward S-parameter spectrum prediction model. The LSTM + Attention network, acting as a forward S-parameter spectrum prediction model, can accurately and efficiently predict the S-parameter spectrum corresponding to the latent variable data and compare it with the real spectrum. In addition, the digits "0" and "1" are used in the design to represent vacuum and metallic materials, respectively, and a 10 x 10 cell array of freestyle metasurface patterns is constructed. The significance of the research method proposed in this paper lies in the following: (1) The freestyle metasurface design significantly expands the possibility of metamaterial design, enabling the creation of diverse metasurface structures that are difficult to achieve with traditional methods. (2) The on-demand design approach can generate high-fidelit

关键词： metasurface variational autoencoder WassersteinGenerative Adversarial Networks S-parameter spectrum prediction inverse metasurface design

来源：评论

学校读者我要写书评

暂无评论

Representing particles' motion patterns in microfluidic imaging platform using deep variational embeddings

引用

JOURNAL OF PHYSICS-PHOTONICS 2025年第3期7卷 035005-035005页

作者： Hong, Tianqi Peng, Meimei Smieja, Marek Fang, Qiyin McMaster Univ Sch Biomed Engn Hamilton ON Canada McMaster Univ Dept Pathol & Mol Med Hamilton ON Canada McMaster Univ Dept Engn Phys Hamilton ON Canada

Understanding the motion properties of cells or particles is important in microfluidic imaging applications. Motion-related analysis has proven to be a valuable tool for phenotyping particulates in biological samples. However, relying solely on trajectory features from individual cells may not always be sufficient to describe their overall motion patterns. This highlights the need for a more effective solution focusing on rotational components in movement. In this study, we developed a generalized motion pattern representation framework using deep variational embeddings to characterize biological samples with different morphology. First, we build a simplified optical setup with sufficient throughput to record sequential frames of cells containing orientational changes. Then, a self-supervised learning pipeline was developed to embed its motion pattern into a latent space. The latent variables are visualized as the generalized motion pattern to represent a sequence of consecutive frames. Finally, segment key frames of individual cells' motion to divide a motion trajectory into consecutive sub-trajectories. Each sub-trajectory has a predefined specific meaning to be collected for downstream motion-related analysis. Our framework has been verified with two cell types with common shapes: plate-like erythrocytes and rod-like yeasts. The results demonstrate that the motion pattern representation is distinct and interpretable for these two samples. Utilized in a motion segmentation application, the represented motion achieved over 90% accuracy with unsupervised clustering, which has significantly enhanced relevant motion analysis. These promising findings underscore the practical value of our developed framework in extracting informative motion patterns for phenotyping.

关键词： motion pattern representation microfluidic imaging variational autoencoder gated recurrent units

来源：评论

学校读者我要写书评

暂无评论

Aerodynamic optimization of supersonic airfoils using bijective cycle generative adversarial networks

引用

THEORETICAL AND APPLIED MECHANICS LETTERS 2025年第4期15卷

作者： Zhao, Chenfei Dai, Yuting Wang, Xue Yang, Chao Huang, Guangjing Beihang Univ Sch Aeronaut Sci & Engn Beijing 100191 Peoples R China Tianmushan Lab Hangzhou 310023 Peoples R China

An efficient, diversified, and low-dimensional airfoil parameterization method is critical to airfoil aerodynamic optimization design. This paper proposes a supersonic airfoil parameterization method based on a bijective cycle generative adversarial network (Bicycle-GAN), whose performance is compared with that of the cVAE-based parameterization method in terms of parsimony, flawlessness, intuitiveness, and physicality. In all four aspects, the Bicycle-GAN-based parameterization method is superior to the cVAE-based parameterization method. Combined with multifidelity Gaussian process regression (MFGPR) surrogate model and a Bayesian optimization algorithm, a Bicycle-GAN-based optimization framework is established for the aerodynamic performance optimization of airfoils immersed in supersonic flow, which is compared with the cVAE-based optimization method in terms of optimized efficiency and effectiveness. The MFGPR surrogate model is established using low-fidelity aerodynamic data obtained from supersonic thin-airfoil theory and high-fidelity aerodynamic data obtained from steady CFD simulation. For both supersonic conditions, the CFD simulation costs are reduced by > 20 % compared with those of the cVAE-based optimization, and better optimization results are obtained through the Bicycle-GAN model. The optimization results for this supersonic flow point to a sharper leading edge, a smaller camber and thickness with a flatter lower surface, and a maximum thickness at 50 % chord length. The advantages of the Bicycle-GAN and MFGPR models are comprehensively demonstrated in terms of airfoil generation characteristics, surrogate model prediction accuracy and optimization efficiency.

关键词： Aerodynamic optimization design Deep learning Generative adversarial network variational autoencoder Multifidelity gaussian process regression

来源：评论

学校读者我要写书评

暂无评论

Advanced temporal deep learning framework for enhanced predictive modeling in industrial treatment systems

RESULTS IN ENGINEERING

引用

RESULTS IN ENGINEERING 2025年 25卷

作者： Ramya, S. Srinath, S. Tuppad, Pushpa JSS Sci & Technol Univ Dept Comp Sci & Engn Mysuru India JSS Sci & Technol Univ Dept Environm Engn Mysuru India

This research introduces an innovative hybrid modeling framework tailored for interval prediction, aimed at forecasting water quality parameters in industrial sewage treatment plants (STPs). It tackles key challenges in the field, including limited data availability, detecting anomalies, and selecting relevant features with precision. By addressing these critical gaps, the study advances predictive analytics for wastewater treatment systems. The main goal was to create a scalable and resilient model that consistently provides accurate forecasts for essential water quality indicators. To accomplish this, variational autoencoders (VAEs) were employed to generate synthetic datasets that mimic real-world patterns, improving data availability and enhancing the model's generalization capabilities. autoencoder paired with a Self-Organizing Map (SOM) was leveraged for anomaly detection and efficient feature selection. The study evaluated advanced architectures, including a Temporal Convolutional Network (TCN), TCN integrated with bidirectional Long Short-Term Memory (BiLSTM), and refined TCN_BiLSTM models featuring preand post-soft attention layers. The final model incorporated Multi- Head Attention mechanisms in both preand post-processing stages (TCN_BiLSTM_MultiHead_Attention), delivering a substantial performance improvement. The TCN_BiLSTM_MultiHead_Attention model proved to be the top performer, delivering state-of-the-art results with R2 scores of 0.9732, 0.9567, and 0.9638 for the training, validation, and test datasets, respectively. On the test set, it achieved an impressive MSE of 0.0008 and an MAE of 0.0198. These results underscore the model's exceptional accuracy in predicting key parameters, including BOD, COD, and AmmoniaNitrogen. The results highlight the significant potential of hybrid deep learning frameworks in capturing temporal patterns and complex dynamics within STP data. By integrating temporal pattern recognition, long-term dependency modeling, and

关键词： Treatment plant variational autoencoder Self-organizing map Temporal convolutional network Bidirectional long short-term memory Gated recurrent units Attention mechanism

来源：评论

学校读者我要写书评

暂无评论

CMATalk: Cross modality alignment for talking head generation

引用

Multimedia Tools and Applications 2025年 1-24页

作者： Cao, Xuan-Nam Trinh, Quoc-Huy Tran, Minh-Triet Faculty of Information Technology University of Science VNU-HCM Ho Chi Minh City Viet Nam Vietnam National University Ho Chi Minh City Viet Nam Department of Computer Science School of Science Aalto University Espoo Finland

Talking head generation has wide applications in virtual assistants, education, and entertainment. Indirect approaches, which leverage facial landmarks as intermediates, offer flexibility in output identity and resolution but often fail to effectively integrate multi-domain information. In this paper, we propose CMATalk, a novel framework that enhances the alignment between audio and facial landmarks within a shared latent space. CMATalk incorporates audio, landmark, and emotional cues through a cross-modal attention mechanism to generate expressive and synchronized facial landmarks. We further introduce a VAE-based image synthesis module that distills knowledge from a pre-trained model to reconstruct high-quality facial images from the predicted landmarks. Extensive experiments demonstrate that our method achieves competitive performance compared to state-of-the-art techniques. Our key contributions include: (1) a latent space alignment strategy for improved audio-landmark correlation, (2) a cross-modal architecture that fuses emotion with audio and landmark features, and (3) a VAE-enhanced decoder for high-fidelity facial reconstruction. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2025.

关键词： Audio-visual integration Landmark prediction Talking head generation variational autoencoder

来源：评论

学校读者我要写书评

暂无评论

Disentangling Reasoning Factors for Natural Language Inference

引用

BIG DATA MINING AND ANALYTICS 2025年第3期8卷 694-711页

作者： Zhou, Xixi Zeng, Limin Zhao, Ziping Bu, Jiajun Liang, Wenjie Wang, Haishuai Zhejiang Univ Affiliated Hosp 1 Dept Radiol Sch Med Hangzhou 310003 Peoples R China Zhejiang Univ Coll Comp Sci Zhejiang Prov Key Lab Serv Robot Hangzhou 310027 Peoples R China Tianjin Normal Univ Coll Comp Sci Tianjin 300387 Peoples R China

Natural Language Inference (NLI) seeks to deduce the relations of two texts: a premise and a hypothesis. These two texts may share similar or different basic contexts, while three distinct reasoning factors emerge in the inference from premise to hypothesis: entailment, neutrality, and contradiction. However, the entanglement of the reasoning factor with the basic context in the learned representation space often complicates the task of NLI models, hindering accurate classification and determination based on the reasoning factors. In this study, drawing inspiration from the successful application of disentangled variational autoencoders in other areas, we separate and extract the reasoning factor from the basic context of NLI data through latent variational inference. Meanwhile, we employ mutual information estimation when optimizing variational autoencoders (VAE)-disentangled reasoning factors further. Leveraging disentanglement optimization in NLI, our proposed a Directed NLI (DNLI) model demonstrates excellent performance compared to state-of-the-art baseline models in experiments on three widely used datasets: Stanford Natural Language Inference (SNLI), Multi-genre Natural Language Inference (MNLI), and Adversarial Natural Language Inference (ANLI). It particularly achieves the best average validation scores, showing significant improvements over the second-best models. Notably, our approach effectively addresses the interpretability challenges commonly encountered in NLI methods.

关键词： Natural Language Processing (NLP) Natural Language Inference (NLI) disentangled representation learning variational autoencoder

来源：评论

学校读者我要写书评

暂无评论

Deep Reinforced Segment Selection and Equalization for Task-Oriented Semantic Communication

引用

IEEE COMMUNICATIONS LETTERS 2024年第8期28卷 1865-1869页

作者： Seon, Joonho Lee, Seongwoo Kim, Jinwook Kim, Soo Hyun Sun, Young Ghyu Seo, Hyowoon Kim, Dong In Kim, Jin Young Kwangwoon Univ Dept Elect Convergence Engn Seoul 01897 South Korea Kwangwoon Univ Dept Elect & Commun Engn Seoul 01897 South Korea Sungkyunkwan Univ Dept Elect & Comp Engn Suwon 16419 South Korea

In this letter, we propose a novel deep reinforcement learning (DRL)-based segment selection and channel equalization strategy for a task-oriented semantic communication (TOSC) system. In non-linear channel conditions, the TOSC framework aims to coordinate computing complexity with task-oriented accuracy. The proposed method navigates this challenge by deploying a DRL agent at the transmitter to eliminate task-irrelevant data and reduce computational complexities while placing a paired DRL agent at the receiver to select an optimal channel equalizer to ensure high accuracy. The simulation results confirm that the proposed system can reduce computational complexity and improve accuracy by 16% over state-of-the-art methods.

关键词： Image segmentation Semantics Task analysis Transmitters Accuracy Receivers Equalizers Semantic communication task-oriented communication reinforcement learning variational autoencoder

来源：评论

学校读者我要写书评

暂无评论

HN3S: A Federated autoencoder framework for Collaborative Filtering via Hybrid Negative Sampling and Secret Sharing

引用

INFORMATION PROCESSING & MANAGEMENT 2024年第2期61卷

作者： Zhang, Lu Li, Guohui Yuan, Ling Ding, Xuanang Rong, Qian Huazhong Univ Sci & Technol Sch Cyber Sci & Engn Wuhan 430074 Peoples R China Huazhong Univ Sci & Technol Sch Software Engn Wuhan 430074 Peoples R China Huazhong Univ Sci & Technol Sch Comp Sci & Technol Wuhan 430074 Peoples R China

Federated recommender systems can serve users with suitable item recommendations while preserving their privacy, but most current works cannot serve non-participant users. We consider that Federated autoencoder for Collaborative Filtering (FAE-CF) is expected to solve this problem. However, FAE-CF encounters privacy protection concerns and high communication overhead issues. In this paper, we investigate FAE-CF's privacy protection capabilities and communication volume, unveiling its potential privacy security issues and inferior practicality. By analyzing the FAE-CF gradient, we propose an effective attack that could recover most user data, with all recovery metrics above 95.53% in our simulation experiments. To address these challenges, we propose Hybrid Negative Sampling and Secret Sharing (HN3S), a privacy preserving method designed for FAE-CF. HN3S uses hybrid negative sampling to reduce communication overhead and secret sharing to protect user data. Experiments conducted on various public datasets with multiple AE-CFs have shown that HN3S can effectively protect the privacy of FAE-CF, reducing the communication overhead by up to 60.44%, while maintaining the performance reduction within 10.78%. Our research reveals significant privacy risks and poor utility inherent in current FAE-CFs, and provides a feasible solution to protect user privacy and balance communication overhead with algorithm performance.

关键词： Federated recommendation Collaborative filtering variational autoencoder Negative sampling Secret sharing

来源：评论

学校读者我要写书评

暂无评论

Robust hierarchical anomaly detection using feature impact in IoT networks

引用

ICT EXPRESS 2025年第2期11卷 358-363页

作者： Rheey, Joohong Park, Hyunggon Ewha Womans Univ Dept Elect & Elect Engn Grad Program Smart Factory Seoul South Korea

Security threats in Internet of Things (IoT) networks increased, but the lack of labeled data and limited resources hinder intrusion detection system design for IoT networks. We propose a robust hierarchical anomaly detection method based on a variational autoencoder for IoT networks. Our proposed approach includes a shallow detection stage for obvious outliers with an in-depth detection stage that explicitly measures the impact of individual features on latent representations using Shapley values, enhancing the ability to detect adversarial attacks without adversarial training. Simulations confirm the effectiveness against adversarial attacks, with almost 100% detection rates for NSL-KDD and CIC-IDS2017 datasets.

关键词： Adversarial attack Anomaly detection Feature impact Intrusion detection system variational autoencoder

来源：评论

学校读者我要写书评

暂无评论

Unsupervised Imputation of Non-Ignorably Missing Data Using Importance-Weighted autoencoders

引用

STATISTICS IN BIOPHARMACEUTICAL RESEARCH 2025年第2期17卷 222-234页

作者： Lim, David K. Rashid, Naim U. Oliva, Junier B. Ibrahim, Joseph G. Univ North Carolina Chapel Hill Dept Biostat McGavran Greenberg Hall 3101 Chapel Hill NC 27599 USA Univ North Carolina Chapel Hill Dept Comp Sci Chapel Hill NC USA

Deep Learning (DL) methods have dramatically increased in popularity in recent years. While its initial success was demonstrated in the classification and manipulation of image data, there has been significant growth in the application of DL methods to problems in the biomedical sciences. However, the greater prevalence and complexity of missing data in biomedical datasets present significant challenges for DL methods. Here, we provide a formal treatment of missing data in the context of variational autoencoders (VAEs), a popular unsupervised DL architecture commonly used for dimension reduction, imputation, and learning latent representations of complex data. We propose a new VAE architecture, NIMIWAE, that is one of the first to flexibly account for both ignorable and non-ignorable patterns of missingness in input features at training time. Following training, samples can be drawn from the approximate posterior distribution of the missing data can be used for multiple imputation, facilitating downstream analyses on high dimensional incomplete datasets. We demonstrate through statistical simulation that our method outperforms existing approaches for unsupervised learning tasks and imputation accuracy. We conclude with a case study of an EHR dataset pertaining to 12,000 ICU patients containing a large number of diagnostic measurements and clinical outcomes, where many features are only partially observed.

关键词： Deep learning Electronic health records Missing data MNAR variational autoencoder

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：