检索结果-内蒙古大学图书馆

Targeted Adversarial Examples Generating Method Based on cVAE in Black Box Settings

Chinese Journal of Electronics 2021年第5期30卷 866-875页

作者： YU Tingyue WANG Shen ZHANG Chunrui WANG Zhenbang LI Yetian YU Xiangzhan School of Cyberspace Science Harbin Institute of Technology Institute of Computer Application China Academy of Engineering Physics State Grid Heilongjiang Power Co. Ltd.

In recent years, adversarial examples has become one of the most important security threats in deep learning applications. For testing the security of deep learning models in adversarial environment,many researches focus on generating adversarial examples quickly and efficiently. In order to solve the problems of existing generative adversarial networks based methods which can not effectively generate the targeted adversarial examples in black box settings, and to improve the temporal performance of gradient-based generating methods,an adversarial examples generating method based on conditional variational autoencoder(cVAE) is proposed in this paper, where a cVAE is designed elaborately to generate adversarial examples without most of the detailed information about the attacked deep learning models, of which the output can be controlled arbitrarily by these crafted inputs, used to test the robustness of deep learning models against adversarial examples. The experimental results show that the proposed method can achieve a comparable attack success rate and a better temporal performance than the existing gradient-based generating methods in black box environment.

关键词： adversarial environment targeted adversarial examples generating method black box settings conditional variational autoencoder deep learning (artificial intelligence) security of data gradient-based generating methods deep learning applications cVAE generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

Enhancement of Track Damage Identification by Data Fusion of Vibration-Based Image Representation

引用

JOURNAL OF NONDESTRUCTIVE EVALUATION 2024年第1期43卷 15-15页

作者： Wang, Shaohua Tang, Lihua Dou, Yinling Li, Zhaoyu Aw, Kean C. Univ Auckland Dept Mech & Mechatron Engn 5 Grafton Rd Auckland 1010 New Zealand Chengdu Vocat & Tech Coll Ind Chengdu 610031 Peoples R China China Railway Eryuan Engn Grp Co Ltd Inst Sci & Technol Chengdu 610031 Peoples R China

In this paper, the vibration-based image representation and data fusion demonstrates distinctive benefit in feature extraction, yielding superior performance for damage identification in railway engineering. Specifically, based on vehicle-track coupled dynamics, the rail vibration datasets under diverse fastener damage conditions are generated. By converting 1-D vibration signals into 2-D grayscale images with recurrence plots (RPs) and the aid of conditional variational autoencoder (CVAE), the acceleration RPs and displacement RPs are fused for enhancing feature extraction. It is demonstrated that detecting the variation in texture patterns and color distribution of the vibration-based images facilitates effective damage identification, mitigating the sensitivity of damage recognition to the deterioration of track irregularity. The results show that the displacement RPs characterised by quasi-static features are more suitable for fastener damage identification. Further, by employing the data fusion that combines both the random dynamic features of the acceleration RPs and quasi-static features of the displacement RPs, the tolerance of measurement range for accurate fastener damage identification can be extended. The robustness of the proposed method is validated after testing different sampling frequencies and additional noise.

关键词： Rail track Damage identification Vibration-based image representation Data fusion Recurrence plot conditional variational autoencoder

来源：评论

学校读者我要写书评

暂无评论

Domain Reconstruction and Resampling for Robust Salient Object Detection 22

Domain Reconstruction and Resampling for Robust Salient Obje...

引用

30th ACM International Conference on Multimedia (MM)

作者： Yan, Senbo Peng, Liang Yu, Chuer Yang, Zheng Liu, Haifeng Cai, Deng Zhejiang Univ State Key Lab CAD & CG Hangzhou Zhejiang Peoples R China FABU Inc Hangzhou Zhejiang Peoples R China

ISBN: (纸本)9781450392037

Salient Object Detection (SOD) aims at detecting the salient objects covering the whole natural scene. However, one of the main problems in SOD is data bias. Natural scenes vary greatly, while each image in the SOD dataset contains a specific scene. It means that each image is just a sampling point in a specific scene, which is not representative and causes serious sampling bias. Building larger datasets is one solution but costly to address the sampling bias. Our method regards the data distribution of natural scenes as a Gaussian Mixture Distribution, and each scene follows a sub-Gaussian distribution. Our main idea is to reconstruct the data distribution of each scene from the sampling images and then resample from the distribution domain. We represent a scene by a distribution instead of a fixed sampling image to reserve the sampling uncertainty in SOD. Specifically, we employ a Style conditional variational autoencoder (Style-CVAE) to reconstruct the data distribution from image styles and a Gaussian Randomize Attribute Filter (GRAF) to reconstruct data distribution from image attributes (such as lightness, saturation, hue, etc.). We resample the reconstructed data distribution according to the Gaussian probability density function and train the SOD model. Experimental results prove that our method outperforms 16 state-of-the-art methods on five benchmarks.

关键词： Salient Object Detection Domain Reconstruction conditional variational autoencoder

来源：评论

学校读者我要写书评

暂无评论

Learning a Generative Motion Model From Image Sequences Based on a Latent Motion Matrix

引用

IEEE TRANSACTIONS ON MEDICAL IMAGING 2021年第5期40卷 1405-1416页

作者： Krebs, Julian Delingette, Herve Ayache, Nicholas Mansi, Tommaso Univ Cote Azur Epione Team INRIA F-06902 Sophia Antipolis France Siemens Healthineers Digital Technol & Innovat Princeton NJ 08540 USA

We propose to learn a probabilistic motion model from a sequence of images for spatio-temporal registration. Our model encodes motion in a low-dimensional probabilistic space - the motion matrix - which enables various motion analysis tasks such as simulation and interpolation of realistic motion patterns allowing for faster data acquisition and data augmentation. More precisely, the motion matrix allows to transport the recovered motion from one subject to another simulating for example a pathological motion in a healthy subject without the need for inter-subject registration. The method is based on a conditional latent variable model that is trained using amortized variational inference. This unsupervised generative model follows a novel multivariate Gaussian process prior and is applied within a temporal convolutional network which leads to a diffeomorphic motion model. Temporal consistency and generalizability is further improved by applying a temporal dropout training scheme. Applied to cardiac cine-MRI sequences, we show improved registration accuracy and spatio-temporally smoother deformations compared to three state-of-the-art registration algorithms. Besides, we demonstrate the model's applicability for motion analysis, simulation and super-resolution by an improved motion reconstruction from sequences with missing frames compared to linear and cubic interpolation.

关键词： Strain Deformable models Probabilistic logic Tracking Image sequences Gaussian processes Data models Motion model deformable registration conditional variational autoencoder Gaussian process latent variable model motion interpolation motion simulation tracking

来源：评论

学校读者我要写书评

暂无评论

Introducing synthetic pseudo panels: application to transport behaviour dynamics

引用

TRANSPORTATION 2021年第5期48卷 2493-2520页

作者： Borysov, Stanislav S. Rich, Jeppe Tech Univ Denmark DTU Dept Technol Management & Econ DK-2800 Lyngby Denmark

In this paper, a method to study travel behaviour dynamics by constructing detailed synthetic pseudo panels from repeated cross-sectional data is presented. The method is based on the modelling of a high-dimensional joint distribution of travel preferences conditional on detailed socio-economic profiles by using a conditional variational autoencoder (CVAE). The CVAE is a neural-network-based generative model which allows the modelling of very detailed joint and conditional distributions, potentially defined by dozens or even hundreds of attributes in a flexible non-parametric form. The proposed method is used to rank detailed cohorts of individuals into slow and fast movers with respect to the speed at which their travel behaviour change over time. This gives an interesting insight into the types of individuals who are easily motivated to change their behaviour as opposed to those who are less flexible. Specifically, we investigate the dynamics of transport preferences for a fixed pseudo panel of individuals from a large Danish cross-sectional data set covering the period from 2006 to 2016. The comparison of the travel preference distributions from 2006 and 2016 shows that the prototypical fast mover is a single young woman who lives in a large city, whereas the typical slow mover is a middle-aged man with high income from a nuclear family who lives in a detached house outside a city. However, given that it is possible to rank individuals across very detailed socio-economic classifications, many other relationships can be explored. Finally, the CVAE can be directly applied to the population synthesis problem in microsimulation by modelling the distribution of socio-economic profiles conditional on other variables.

关键词： Dynamics of behaviour Pseudo-panel analysis conditional variational autoencoder Deep learning Generative modelling Population synthesis

来源：评论

学校读者我要写书评

暂无评论

Development of a surrogate model to improve the efficiency of groundwater level fluctuation pattern-based hydrologic properties evaluation

引用

JOURNAL OF HYDROLOGY 2023年第1期619卷

作者： Jeong, Jiho Jeong, Jina Kyungpook Natl Univ Dept Geol Daegu South Korea

In this study, a method of aquifer hydrologic property estimation incorporating the deep learning method was developed to improve the estimation efficiency of a process-based model based on groundwater level fluctuation (GLF) patterns. As a reference study, a data-driven method suggested by Jeong et al. (2020) was considered;the uncertainty of the GLF patterns resulting from different yearly patterns of precipitation, which were considered as noise in the previous study, was effectively discarded using the newly proposed method of applying the conditional variational autoencoder (CVAE). The CVAE was used to acquire the specific GLF patterns under certain identical precipitation patterns for all the monitoring stations. The data-driven hydrologic property estimation model was developed to predict two hydrologic parameters (rho and k) of the process-based model using the generated GLF patterns from the CVAE network as the input variables. The actual GLF and precipitation data that were acquired from nationwide groundwater monitoring stations in South Korea were applied to validate the developed method. It was found that the estimated and target hydrologic properties were highly correlated (correlation coefficients [CC]: 0.9833 and 0.9589 for rho and k, respectively), which significantly improved the results when compared to the previous study (CC: 0.7207 and 0.8663 for alpha/n and k, respectively). Consequently, the developed model can contribute to a more accurate hydrologic property estimation of aquifers. Additionally, it can facilitate efficient groundwater development planning since the manual fitting of the process-based model by an expert is not required.

关键词： Hydrologic property estimation Data-driven estimation model Groundwater level fluctuation pattern Precipitation pattern conditional variational autoencoder

来源：评论

学校读者我要写书评

暂无评论

Few-shot fatigue damage evaluation of aircraft structure using neural augmentation and deep transfer learning

引用

ENGINEERING FAILURE ANALYSIS 2023年 148卷

作者： Che, Changchang Wang, Huawei Xiong, Minglan Luo, Shici Nanjing Forestry Univ Coll Automobile & Traff Engn Nanjing Peoples R China Nanjing Univ Aeronaut & Astronaut Coll Civil Aviat Nanjing Peoples R China

To solve the problems of few-shot samples, different structural degradation trends and poor damage evaluation effect in fatigue damage evaluation of aircraft structure, an intelligent eval-uation method based on neural augmentation and deep transfer learning (NA-DTL) is proposed in this paper. Firstly, the fatigue damage is divided into three risk levels according to the length of crack, and conditional variational autoencoder (CVAE) and one-dimensional convolutional neural network (1-DCNN) are constructed to form the neural augmentation model for collaborative optimization of augmentation network and classification network. Subsequently, CVAE is used to generate massive fatigue damage samples, which can provide data support for building of crack length evaluation model. In addition, model-based transfer learning method is applied for damage evaluation according to the trained 1-DCNN. The fatigue crack growth dataset of aircraft aluminum lap joint is utilized to verify the effectiveness of the proposed method. The results show that the proposed method can achieve more accurate evaluation results compared with other models.

关键词： Fatigue damage evaluation Few-shot learning conditional variational autoencoder Neural augmentation Deep transfer learning

来源：评论

学校读者我要写书评

暂无评论

Diversifying Emotional Dialogue Generation via Selective Adversarial Training

引用

SENSORS 2023年第13期23卷 5904-5904页

作者： Li, Bo Zhao, Huan Zhang, Zixing Hunan Univ Coll Comp Sci & Elect Engn Changsha 410082 Peoples R China

Emotional perception and expression are very important for building intelligent conversational systems that are human-like and attractive. Although deep neural approaches have made great progress in the field of conversation generation, there is still a lot of room for research on how to guide systems in generating responses with appropriate emotions. Meanwhile, the problem of systems' tendency to generate high-frequency universal responses remains largely unsolved. To solve this problem, we propose a method to generate diverse emotional responses through selective perturbation. Our model includes a selective word perturbation module and a global emotion control module. The former is used to introduce disturbance factors into the generated responses and enhance their expression diversity. The latter maintains the coherence of the response by limiting the emotional distribution of the response and preventing excessive deviation of emotion and meaning. Experiments are designed on two datasets, and corresponding results show that our model outperforms existing baselines in terms of emotional expression and response diversity.

关键词： dialog systems emotional response generation diversity enhancement latent variables conditional variational autoencoder

来源：评论

学校读者我要写书评

暂无评论

Deep neural network-based reduced-order modeling of ion-surface interactions combined with molecular dynamics simulation

引用

JOURNAL OF PHYSICS D-APPLIED PHYSICS 2023年第38期56卷 384005-384005页

作者： Kim, Byungjo Bae, Jinkyu Jeong, Hyunhak Hahn, Seung Ho Yoo, Suyoung Nam, Sang Ki Samsung Elect Co Ltd Mechatron Res Suwon 18448 Gyeonggi Do South Korea

With the advent of complex and sophisticated architectures in semiconductor device manufacturing, atomic-resolution accuracy and precision are commonly required for industrial plasma processing. This demands a comprehensive understanding of the plasma-material interactions-particularly for forming fine high-aspect ratio (HAR) feature patterns with sufficiently high yield in wafer-level processes. In particular, because the shape distortion in HAR pattern etching is attributed to the deviation of the energetic ion trajectory, the detailed ion-surface interactions need to be thoroughly investigated. In this study, molecular dynamics (MD) simulations were utilized to obtain a fundamental understanding of the collisional nature of accelerated Ar ions on the fluorinated Si surface that may appear on the sidewall of the HAR etched hole. High-fidelity data for ion-surface interaction features representing the energy and angle distributions (EADs) of sputtered atoms for varying degrees of surface F coverage and ion incident angles were obtained via extensive MD simulations. A deep learning-based reduced-order modeling (DL-ROM) framework was developed for efficiently predicting the characteristics of the ion-surface interactions. In the ROM framework, a conditional variational autoencoder (AE) was implemented to obtain regularized latent representations of the distributional data with the condition of the governing factors of the physical system. The proposed ROM framework accurately reproduced the MD simulation results and significantly outperformed various DL-ROMs, such as AE, sparse AE, contractive AE, denoising AE, and variational AE. From the inferred features of the sputtering yield and EADs of sputtered/scattered species, significant insights can be obtained regarding the ion interactions with the fluorinated surface. As the ion incident angle deviated from the glancing-angle range (incident angle >80 & DEG;), diffuse reflection behavior was observed, which can substa

关键词： plasma-material interaction molecular dynamics deep learning reduced-order modeling conditional variational autoencoder

来源：评论

学校读者我要写书评

暂无评论

Unsupervised pathology detection in medical images using conditional variational autoencoders

引用

INTERNATIONAL JOURNAL OF COMPUTER ASSISTED RADIOLOGY AND SURGERY 2019年第3期14卷 451-461页

作者： Uzunova, Hristina Schultz, Sandra Handels, Heinz Ehrhardt, Jan Univ Lubeck Inst Med Informat Lubeck Germany

PurposePathology detection in medical image data is an important but a rather complicated task. In particular, the big variability of the pathologies is a challenge to automatic detection methods and even to machine learning methods. Supervised algorithms would usually learn the appearance of a single pathological structure based on a large annotated dataset. As such data is not usually available, especially in large amounts, in this work we pursue a different unsupervised *** method is based on learning the entire variability of healthy data and detect pathologies by their differences to the learned norm. For this purpose, we use conditional variational autoencoders which learn the reconstruction and encoding distribution of healthy images and also have the ability to integrate certain prior knowledge about the data (condition).ResultsOur experiments on different 2D and 3D datasets show that the approach is suitable for the detection of pathologies and deliver reasonable Dice coefficients and AUCs. Also this method can estimate missing correspondences in pathological images and thus can be used as a pre-step to a registration method. Our experiments show improving registration results on pathological data when using this *** the presented approach is suitable for a rough pathology detection in medical images and can be successfully used as a preprocessing step to other image processing methods.

关键词： conditional variational autoencoder Unsupervised pathology detection Image registration

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：