检索结果-内蒙古大学图书馆

Towards the AlexNet Moment for Homomorphic Encryption: HCNN, the First Homomorphic CNN on Encrypted Data With GPUs

IEEE TRANSACTIONS ON EMERGING TOPICS IN COMPUTING 2021年第3期9卷 1330-1343页

作者： Al Badawi, Ahmad Jin, Chao Lin, Jie Mun, Chan Fook Jie, Sim Jun Tan, Benjamin Hong Meng Nan, Xiao Aung, Khin Mi Mi Chandrasekhar, Vijay Ramaseshan ASTAR Inst Infocomm Res I2R Singapore 138632 Singapore

Deep Learning as a Service (DLaaS) stands as a promising solution for cloud-based inference applications. In this setting, the cloud has a pre-learned model whereas the user has samples on which she wants to run the model. The biggest concern with DLaaS is the user privacy if the input samples are sensitive data. We provide here an efficient privacy-preserving system by employing high-end technologies such as Fully Homomorphic Encryption (FHE), Convolutional neural networks (CNNs) and Graphics processing Units (GPUs). FHE, with its widely-known feature of computing on encrypted data, empowers a wide range of privacy-concerned applications. This comes at high cost as it requires enormous computing power. In this article, we show how to accelerate the performance of running CNNs on encrypted data with GPUs. We evaluated two CNNs to classify homomorphically the MNIST and CIFAR-10 datasets. Our solution achieved sufficient security level (> 80 bit) and reasonable classification accuracy (99) and (77.55 percent) for MNIST and CIFAR-10, respectively. In terms of latency, we could classify an image in 5.16 seconds and 304.43 seconds for MNIST and CIFAR-10, respectively. Our system can also classify a batch of images (> 8,000) without extra overhead.

关键词： Servers Encryption Computational modeling artificial neural networks Training Deep learning privacy-preserving technologies homomorphic encryption implementation GPUs

来源：评论

学校读者我要写书评

暂无评论

An Efficient Face Recognition Method Based on CNN

An Efficient Face Recognition Method Based on CNN

引用

Power Electronics, Computer applications (ICPECA), IEEE International Conference on

作者： Xiu He Feng Ding Guangzhou Xinhua University Guangzhou China

As a prevailing research in artificial intelligence, the application of computer vision is widely used in many fields which are closely related to people's livelihood, such as industrial automation, new retail industry, smart transportation and security monitoring. And the proposed face recognition method is a branch in the field of computer vision, it integrates neural networks, biology, image signal processing, machine learning and other fields, which promote research and cross-development among different disciplines. Hence, this paper focuses on face recognition method by using convolutional neural network(CNN), and CNN has the property of "weight sharing", which has been widely popularized in image recognition, it can greatly simplify the work of large-scale network training. The experiments demonstrate that the proposed face recognition method is successful, and the accuracy of the proposed method can be as high as 98%.

关键词： Training Computer vision image recognition image color analysis Face recognition Color Signal processing

来源：评论

学校读者我要写书评

暂无评论

Development of precise forgery detection algorithms in digital radiography images using convolution neural network

引用

APPLIED SOFT COMPUTING 2023年 138卷

作者： El Tokhy, Mohamed S. Egyptian Atom Energy Author Engn Dept NRC Cairo Egypt

The widespread availability of forged image software necessitates the integrity verification of digital images in industrial and medical applications. Because of image manipulation, detecting small tampering and duplicated forgery from digital radiography (gamma and x-ray) images has become a research challenge, Two essential approaches are proposed for forgery detection from digital radiography images. A precise forgery detection approach with pretrained deep convolution neural networks (CNN) is conducted. Alexnet, Resnet-18 and VGG-19 are three pretrained networks for features extraction. artificial neural network (ANN) and multiclass support vector machine (MSVM) classifiers are applied for classifying the extracted features into authentic or forged. The second suggested approach depends on Haralick and Zoning extractors. These extracted features are trained and tested using the K-nearest neighbors (KNN) classifier. The suggested approaches are investigated using several manipulated industrial (gamma welding images) and medical (spine images) datasets images. Besides, these approaches are tested with several color benchmark dataset images. The results are verified using a variety of evaluation metrics. The approaches are validated through comparison with published work and high agreements are demonstrated. For digital radiography images, Alexnet pretrained network with MSVM, Resnet-18 pretrained network with ANN and Haralick extractor with KNN achieve the highest accuracy and assessment metrics. It is observed that the performance of pretrained CNN outperforms that of conventional classification algorithms in respect of accuracy with computational time. The developed approaches allow for the precise detection of forgery regions in x-ray and gamma radiographic images as well as digital images.& COPY;2023 Elsevier B.V. All rights reserved.

关键词： Algorithms Gamma radiography Nuclear forensic Digital image processing

来源：评论

学校读者我要写书评

暂无评论

SELECTIVE LOSSY image COMPRESSION FOR AUTONOMOUS SYSTEMS 22

SELECTIVE LOSSY IMAGE COMPRESSION FOR AUTONOMOUS SYSTEMS

引用

23rd Symposium on image, Signal processing and artificial Vision (STSIVA)

作者： Sood, Shreyan Ahuja, Yatharth Delhi Technol Univ Dept Appl Math New Delhi India Delhi Technol Univ Dept Elect Engn New Delhi India

ISBN: (纸本)9781665416696

The main objective of this paper was to effectively interface object detection based on Convolution neural networks (CNNs) with selective lossy image compression techniques to improve the efficiency of subsequent image operations and reduce the memory requirement for storing the images in autonomous applications of self-driving vehicles. Object detection and localization was performed using 2 state-of-the-art CNN based models from the Tensorflow 2.0 Object Detection API - Faster R-CNN ResNet152 V1 1024x1024 and CenterNet HourGlass104 1024x1024. Lossy image Compression centred around the most prominent detected object (which is preserved) is done through 3 techniques - K-Means Clustering (KM), Genetic Algorithm (GA), Discrete Cosine Transform (DCT). The compressed and preserved parts were recombined to produce the final image. Analysis of the results obtained from different models and compression techniques was carried out. It was found that DCT produced the best results on both the models.

关键词： CNN Object Detection Lossy image Compression Autonomous Systems

来源：评论

学校读者我要写书评

暂无评论

Exploring adversarial robustness of JPEG AI: methodology, comparison and new methods

arXiv

引用

arXiv 2024年

作者： Kovalev, Egor Bychkov, Georgii Abud, Khaled Gushchin, Aleksandr Chistyakova, Anna Lavrushkin, Sergey Vatolin, Dmitriy Antsiferova, Anastasia MSU Institute for Artificial Intelligence ISP RAS Research Center for Trusted Artificial Intelligence Lomonosov Moscow State University Russia Laboratory of Innovative Technologies for Processing Video Content Innopolis University Russia

Adversarial robustness of neural networks is an increasingly important area of research, combining studies on computer vision models, large language models (LLMs), and others. With the release of JPEG AI — the first standard for end-to-end neural image compression (NIC) methods — the question of its robustness has become critically significant. JPEG AI is among the first international, real-world applications of neural-network-based models to be embedded in consumer devices. However, research on NIC robustness has been limited to open-source codecs and a narrow range of attacks. This paper proposes a new methodology for measuring NIC robustness to adversarial attacks. We present the first large-scale evaluation of JPEG AI’s robustness, comparing it with other NIC models. Our evaluation results and code are publicly available online (link is hidden for a blind review). Copyright © 2024, The Authors. All rights reserved.

关键词： neural network models

来源：评论

学校读者我要写书评

暂无评论

Convolutional neural Network (CNN)-Based Measurement of Properties in Liquid-Liquid Systems

引用

PROCESSES 2023年第5期11卷

作者： Neuendorf, Laura Mueller, Pascal Lammers, Keno Kockmann, Norbert TU Dortmund Univ Dept Biochem & Chem Engn Lab Equipment Design Emil Figge Str 68 D-44227 Dortmund Germany

The rise of artificial intelligence (AI)-based image analysis has led to novel application possibilities in the field of solvent analytics. Using convolutional neural networks (CNNs), better and more automated analysis of optically visible phenomena becomes feasible, broadening the spectrum of non-invasive measurements. These so-called smart sensors have attracted increasing attention in pharmaceutical and chemical process engineering;their additional sensor data enables more precise process control as additional process parameters can be monitored. This contribution presents an approach to analyzing single rising droplets to determine their physical properties;for example, geometrical parameters such as diameter, projection area and volume. Additionally, the rising velocity is determined, as well as the density and interfacial tension of the rising liquid droplet, determined from the force balance. Thus, a method was developed for analyzing liquid-liquid properties suitable for real-time applications. Here, the size range of the investigated droplet diameters lies between 0.68 mm and 7 mm with an accuracy for AI detecting droplets of +/- 4 mu m. The obtained densities lie between 0.822 kg.m (-3) for rising n-butanol droplets and 0.894 kg.m (-3) for toluene droplets. For the derived parameters, such as the interfacial tension estimation, all of the data points lie in a range from 12.75 mN.m (-1) to 15.25 mN.m (-1). The trueness of the investigated system thus is in a range from -1 to +0.4 mN.m (-1), with a precision of +/- 0.3 to +/- 0.6 mN.m (-1). For density estimation using our system, a standard deviation of 1.4 kg m 3 from the literature was determined. Using camera images in conjunction with image analysis improved by artificial intelligence algorithms, combined with using empirical mathematical formulas, this article contributes to the development of easily accessible, cheap sensors.

关键词： convolutional neural networks image processing visual sensors tensiometer densitometer density estimation interfacial tension multiphase flow single rising droplets

来源：评论

学校读者我要写书评

暂无评论

Research on Multi-Label Classification of Chinese Text Based on Word-Label Probability and DistilBERT

Research on Multi-Label Classification of Chinese Text Based...

引用

image processing and Computer applications (ICIPCA), IEEE International Conference on

作者： Hong Zhao Weiquan Tian Ding An Shan Mi School of Computer Science and Communication Lanzhou University of Technology Lanzhou China CCCC Big Data Technology Co. Ltd. Lanzhou China

ISBN: (数字)9798350360240

ISBN: (纸本)9798350384161

Multi-label text classification is a fundamental task in the field of natural language processing. Currently, there are issues in the Chinese multi-label text classification tasks, such as insufficient extraction of text label features and a lack of learning about the correlations between labels. To address these challenges, we proposed a Multi-label Text Classification Model based on DistilBERT and Word-Label Probabilities (DBWL). The model consists of a metric space module that integrates word-label probabilities and a deep neural network module based on DistilBERT. Firstly, the trained Labeled Latent Dirichlet Allocation (Labeled-LDA) model is utilized to obtain the probabilities of words associated with labels, which are then used to calculate the text mapping vectors and map them into the same metric space as the labels. Similarity measurement is employed to extract features between text-label pairs and label-label pairs. Then, in the deep neural network module, DistilBERT is employed to jointly embed texts and labels, capturing rich semantic information about text labels, followed by feature extraction using TextCNN. Finally, the features extracted from both modules are weighted and fused for label prediction. Experimental results on two benchmark datasets demonstrate that the proposed model outperforms current mainstream multi-label classification models in terms of key performance metrics.

关键词： Computational modeling Text categorization Semantics artificial neural networks Probability Feature extraction Extraterrestrial measurements Vectors Natural language processing Resource management

来源：评论

学校读者我要写书评

暂无评论

Classification and regression of trajectories rendered as images via 2D Convolutional neural networks

arXiv

引用

arXiv 2024年

作者： Nicolai, Mariaclaudia Cabini, Raffaella Fiamma Pizzagalli, Diego Ulisse Euler institute Univeristà della Svizzera italiana Lugano Switzerland University of Pavia Pavia Italy

Trajectories can be regarded as time-series of coordinates, typically arising from motile objects. Methods for trajectory classification are particularly important to detect different movement patterns, while methods for regression to compute motility metrics and forecasting. Recent advances in computer vision have facilitated the processing of trajectories rendered as images via artificial neural networks with 2d convolutional layers (CNNs). This approach leverages the capability of CNNs to learn spatial hierarchies of features from images, necessary to recognize complex shapes. Moreover, it overcomes the limitation of other machine learning methods that require input trajectories with a fixed number of points. However, rendering trajectories as images can introduce poorly investigated artifacts such as information loss due to the plotting of coordinates on a discrete grid, and spectral changes due to line thickness and aliasing. In this study, we investigate the effectiveness of CNNs for solving classification and regression problems from synthetic trajectories that have been rendered as images using different modalities. The parameters considered in this study include line thickness, image resolution, usage of motion history (color-coding of the temporal component) and anti-aliasing. Results highlight the importance of choosing an appropriate image resolution according to model depth and motion history in applications where movement direction is critical. © 2024, CC BY.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Implementation and Analysis of Sparse DNN on GPU

Implementation and Analysis of Sparse DNN on GPU

引用

IEEE International Symposium on Smart Electronic Systems (iSES)

作者： Aparna Nair M K Nandhu Sam Minu A Pillai Nalesh S Kala S Department of ECE Indian Institute of Information Technology Kottayam Kerala India Institute for Integrated Programmes and Research in Basic Sciences Mahatma Gandhi University Kerala India Department of Electronics Cochin University of Science and Technology Kerala India

ISBN: (数字)9798331533229

ISBN: (纸本)9798331533236

Deep neural networks are being widely used in applications like object detection, image classification, etc. Having billions of parameters in the networks poses challenges in terms of storage requirements, and computational complexity. Exploiting sparsity in weights and input activations results in reduced memory and accelerated computations. This paper evaluates the performance of a sparse neural network inference on GPU that utilizes a clustering technique to convert the intermediate results into sparser representation during inference time with a sparse matrix storage format. The methodology utilizes Delta-Compressed Sparse Row sparse weight storage format for reduced memory requirements. The proposed approach has been implemented on NVIDIA Tesla T4 GPU, using medium-scale sparse DNN trained with MNIST dataset. Our implementation has achieved an average reduction of 4% in storage requirement and a maximum speedup of 1.92 × compared to baseline SNICIT implementation.

关键词： Embedded systems Single instruction multiple data Memory management Graphics processing units artificial neural networks Object detection Inference algorithms Sparse matrices Computational complexity image classification

来源：评论

学校读者我要写书评

暂无评论

Scalable Construction of Formal Error Guaranteed LUT-Based Approximate Multipliers with Analytical Worst-Case Error Bound 26th

Scalable Construction of Formal Error Guaranteed LUT-Based ...

引用

26th International Symposium on VLSI Design and Test, VDAT 2022

作者： Venkatesh, Anishetti Jha, Chandan Kumar Vinod, G.U. Fujita, Masahiro Singh, Virendra Indian Institute of Technology Bombay Mumbai India German Research Center for Artificial Intelligence Bremen Germany Qualcomm Chennai India University of Tokyo Tokyo Japan

ISBN: (纸本)9783031215131

In approximate circuits, the functionality of the circuit is modified to improve non-functional parameters such as area, power, and delay. These circuits are targeted for error-resilient applications like image processing, neural networks, etc. These applications can achieve acceptable output results even after the introduction of approximation in computations. Error computation becomes crucial in these applications and can be done either using simulation or formal approaches. Both these methods of error computation are not scalable for circuits with inputs larger than 32 bits. In this work, we used scalable construction of large array multipliers using smaller formally worst-case error guaranteed array multipliers obtained using an LUT-SAT based approach and guaranteed overall worst-case error bounds analytically. Our 8 × 8 approximate multipliers based on the LUT-SAT approach have 0.59% less area as compared to the state-of-the-art EvoApprox library-based multipliers. We compared the 16 × 16 multipliers constructed using four 8 × 8 LUT-SAT-based multipliers with 16 × 16 multipliers constructed using four 8 × 8 Evo multipliers. We have performed the area and various error metric comparisons. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： image processing

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

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

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

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

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：