检索结果-内蒙古大学图书馆

FeSHI: Feature Map-Based Stealthy Hardware Intrinsic Attack

IEEE ACCESS 2021年 9卷 115370-115387页

作者： Odetola, Tolulope A. Khalid, Faiq Mohammed, Hawzhin Sandefur, Travis C. Hasan, Syed Rafay Tennessee Technol Univ Dept Elect & Comp Engn Cookeville TN 38505 USA Tech Univ Wien TU Wien Dept Comp Engn A-1040 Vienna Austria

Convolutional neural networks (CNN) have shown impressive performance in computer vision, natural language processing, and many other applications, but they exhibit high computations and substantial memory requirements. To address these limitations, especially in resource-constrained devices, the use of cloud computing for CNNs is becoming more popular. This comes with privacy and latency concerns that have motivated the designers to develop embedded hardware accelerators for CNNs. However, designing a specialized accelerator increases the time-to-market and cost of production. Therefore, to reduce the time-to-market and access to state-of-the-art techniques, CNN hardware mapping and deployment on embedded accelerators are often outsourced to untrusted third parties, which is going to be more prevalent in futuristic artificial intelligence of things (AIoT) systems. These AIoT systems anticipates horizontal collaboration among different resource constrained AIoT node devices, where CNN layers are partitioned and these devices collaboratively compute complex CNN tasks. This horizontal collaboration opens another attack surface to the CNN-based application, like inserting the hardware Trojans (HT) into the embedded accelerators designed for the CNN. Therefore, there is a dire need to explore this attack surface for designing the secure embedded hardware accelerators for CNNs. Towards this goal, in this paper, we exploited this attack surface to propose an HT-based attack called FeSHI. Since in horizontal collaboration of RC AIoT devices different sections of CNN architectures are outsourced to different untrusted third parties, the attacker may not know the input image, but it has access to the layer-by-layer output feature maps information for the assigned sections of the CNN architecture. This attack exploits the statistical distribution, i.e., Gaussian distribution, of the layer-by-layer feature maps of the CNN to design two triggers for stealthy HT with a very low

关键词： Convolutional neural networks IP networks Field programmable gate arrays Trojan horses Statistical analysis Gaussian distribution Collaboration Convolutional neural Network CNN hardware security edge intelligence AIoT FPGA hardware Trojan hardware intrinsic attack

来源：评论

学校读者我要写书评

暂无评论

AI Based Automated image Caption Tool Implementation for Visually Impaired 1

AI Based Automated Image Caption Tool Implementation for Vis...

引用

1st International Conference on Industrial Electronics Research and applications, ICIERA 2021

作者： Wadhwa, Vanshika Gupta, Bhoomi Gupta, Sachin Maharaja Agrasen Institute of Technology Department of It Delhi India School of Engineering and Technology Mvn University Haryana India

ISBN: (纸本)9781665435420

image captioning is a rapidly emerging area in the artificial Intelligence applications for natural language definitions. It works at the confluence of image data obtained through datasets, and the sentence definitions towards capturing meaningful interpretations of the interaction that exists between them. It uses CNN's (Convolutional neural networks) reading techniques in image and LSTM (Long Short Term Memory) type RNN (Recurrent neural Network) over sentences together so that the computer can see the context of the image and display it in a natural language like English. This paper combines the application of computer vision and natural language processing towards building assistive technology that supplements visual data like images by providing braille readable captions for the visually impaired to get a better sense of what is happening around them and understand their surroundings. © 2021 IEEE.

关键词： Computer vision

来源：评论

学校读者我要写书评

暂无评论

Accelerating Weed Detection for Smart Agricultural Sprayers Using a neural processing Unit

SSRN

引用

SSRN 2024年

作者： Herterich, Nils Liu, Kai Stein, Anthony Robert Bosch GmbH Heilbronner Straße 358 Stuttgart70469 Germany University of Hohenheim Department of Artificial Intelligence in Agricultural Engineering Garbenstraße 9 Stuttgart70599 Germany

Smart spraying technology represents a breakthrough in precision agriculture, enabling innovative weed control methods and increasing crop protection effectiveness through the use of deep neural networks (DNNs). This advancement brings its own set of challenges, particularly in terms of the time neural networks need to reach conclusions. In large sprayers, weed detection is performed by multiple sensors, each focused on a distinct region of interest. While existing research has mainly concentrated on powerful GPUs, the shift to practical applications requires the use of energy-efficient embedded systems to enable intelligent spraying in the field. In this study, the effects of different backbone and neck architectures on an object detection model are analyzed in terms of inference time and compatibility with a neural processing Unit (NPU), focusing on efficiency and accuracy. Considering real-time operating conditions, the neural network architectures were optimized for a maximum inference time of 50 ms and a model size of less than 10 MB. To achieve this, the models were validated for operational compatibility during the deployment process and quantized to an 8-bit integer data format. A dataset of 12 weed species was used to test a variety of novel lightweight CNN and hybrid transformer backbones and necks of an object detection model at different image resolutions. The results show that these core components of the model have different effects on inference time and accuracy, with certain configurations meeting the required specifications. The speedup factor, which describes the comparison of CPU and NPU inference time on the target hardware, significantly influences the efficiency. Experimental tests with different backbones showed distinct advantages in weed detection. Building on this success, further efficiency improvements could be achieved by replacing the neck. While the required quantization led to a reduction in accuracy, the proposed combination of compo

关键词： Object detection

来源：评论

学校读者我要写书评

暂无评论

Comparative Analysis of Machine Learning and Deep Learning Algorithms for Plant Disease Classification and Detection

Comparative Analysis of Machine Learning and Deep Learning A...

引用

Advanced Computing & Communication Technologies (ICACCTech), International Conference on

作者： Harminder Kaur Neeraj Raheja Amit Kumar Bindal Department of Computer Science and Engineering Maharishi Markandeshwar Engineering College Maharishi Markandeshwar (Deemed to be University) Mullana Ambala Haryana India

ISBN: (数字)9798331519056

ISBN: (纸本)9798331519063

Agriculture, a crucial foundation for human existence, exists at the intersection of conventional methods and modern innovations. Any country's primary issue in agriculture science is effective plant disease control, driven by the growing need for food due to population growth. Moreover, developments in modern technology have greatly improved the precision and effectiveness of plant detection. A critical challenge in the agricultural sector is leaf diseases, significantly impacting crop production and economic profit. Integrating digitalization and technology in agriculture is imperative for augmenting productivity and ensuring food security. This research emphasizes the principal concerns and obstacles associated with classifying leaf diseases. By experimenting with different approaches, researchers have already achieved great strides in the identification and categorization of illnesses. But due to assessments, new information, and discussions, changes are required. The resilience of deep learning (DL) and machine learning (ML) techniques, such as feed-forward neural networks (FFNN), k-means clustering (KMC), fuzzy logic (FL), artificial neural networks (ANN), and genetic algorithms (GA),so forth, has been demonstrated by earlier research. Due to their inherent ability to automatically obtain relevant visual information and understand spatial hierarchies, CNNs are usually the preferred option for image identification and classification among the ML and DL methods addressed in this study. The selection between machine learning and deep learning is dependent upon numerous issues, the data that is available, and the amount of processing power that is available. Because of this, DL—primarily using CNNs are advised for many complicated image detection and classification applications. They also show outstanding results for classification and detection on their datasets but not on others. Lastly, by using a variety of methods for processing images in the field of artific

关键词： Deep learning Fuzzy logic Support vector machines Plant diseases Visualization Technological innovation Feature extraction Agriculture Data mining Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Research on Fault Diagnosis and Parameter Identification of Bidirectional LLC Resonant Converter Based on Data Driven

Research on Fault Diagnosis and Parameter Identification of ...

引用

image processing, Electronics and Computers (IPEC), Asia-Pacific Conference on

作者： Xiaorui Zhang Shuo Jiang Pingxin Ma Dongyang Luo Tian Zhao Fengping He China Academy of Space Technology(CAST) Institute of Telecommunication and Navigation Satellites Beijing China

ISBN: (数字)9798350374407

ISBN: (纸本)9798350374414

At present, few people have studied the fault diagnosis of bidirectional LLC resonant converters. In order to promote the development of digital and intelligent fault diagnosis of power converters, this article analyzes the structural and parametric fault characteristics of bidirectional LLC resonant converters, designs structural fault diagnosis algorithms based on Support Vector Machine (SVM) and Random Forest (RF), and compares the diagnostic results of the two algorithms. Among them, the accuracy of random forest fault diagnosis is $100 \%$, the accuracy of SVM fault diagnosis is $98.58 \%$, the accuracy of RF fault diagnosis is higher, and the fault diagnosis speed is faster; At the same time, a bidirectional LLC resonant converter resonant capacitor parameter identification algorithm based on Extreme Learning Machines (ELM) and Deep neural Network (DNN) was studied. As a deep learning algorithm, DNN has an identification error of less than $5 \%$ and significantly better performance than ELM. The data-driven bidirectional LLC resonant converter fault diagnosis and parameter identification carried out in this article have achieved good results, laying a foundation for subsequent engineering applications.

关键词： Fault diagnosis Support vector machines Radio frequency Parameter estimation Accuracy Resonant converters artificial neural networks Prediction algorithms Circuit faults Random forests

来源：评论

学校读者我要写书评

暂无评论

Photonic Convolutional neural networks Using Integrated Diffractive Optics

引用

IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS 2020年第5期26卷

作者： Ong, Jun Rong Ooi, Chin Chun Ang, Thomas Y. L. Lim, Soon Thor Png, Ching Eng ASTAR Inst High Performance Comp Singapore 138632 Singapore

With recent rapid advances in photonic integrated circuits, it has been demonstrated that programmable photonic chips can be used to implement artificial neural networks. Convolutional neural networks (CNN) are a class of deep learning methods that have been highly successful in applications such as image classification and speech processing. We present an architecture to implement a photonic CNN using the Fourier transform property of integrated star couplers. We show, in computer simulation, high accuracy image classification using the MNIST dataset. We also model component imperfections in photonic CNN and show that the performance degradation can be recovered in a programmable chip. Our proposed architecture provides a large reduction in physical footprint compared to current implementations as it utilizes the natural advantages of optics and hence offers a scalable pathway towards integrated photonic deep learning processors.

关键词： Photonics Discrete Fourier transforms Couplers Optical waveguides Convolution High-speed optical techniques Optical interferometry artificial neural networks neuromorphics photonic integrated circuits silicon photonics

来源：评论

学校读者我要写书评

暂无评论

AdaBM: On-the-Fly Adaptive Bit Mapping for image Super-Resolution

AdaBM: On-the-Fly Adaptive Bit Mapping for Image Super-Resol...

引用

Conference on Computer Vision and Pattern Recognition (CVPR)

作者： Cheeun Hong Kyoung Mu Lee Dept. of ECE & ASRI IPAI Seoul National University Seoul Korea

ISBN: (数字)9798350353006

ISBN: (纸本)9798350353013

Although image super-resolution (SR) problem has ex-perienced unprecedented restoration accuracy with deep neural networks, it has yet limited versatile applications due to the substantial computational costs. Since differ-ent input images for SR face different restoration difficul-ties, adapting computational costs based on the input image, referred to as adaptive inference, has emerged as a promising solution to compress SR networks. Specifically, adapting the quantization bit-widths has successfully re-duced the inference and memory cost without sacrificing the accuracy. However, despite the benefits of the resul-tant adaptive network, existing works rely on time-intensive quantization-aware training with full access to the origi-nal training pairs to learn the appropriate bit allocation policies, which limits its ubiquitous usage. To this end, we introduce the first on-the-fly adaptive quantization frame-work that accelerates the processing time from hours to sec-onds. We formulate the bit allocation problem with only two bit mapping modules: one to map the input image to the image-wise bit adaptation factor and one to obtain the layer-wise adaptation factors. These bit mappings are cali-brated and fine-tuned using only a small number of calibration images. We achieve competitive performance with the previous adaptive quantization methods, while the processing time is accelerated by × 2000. Codes are available at https://***/Cheeun/AdaBM.

关键词： Training Quantization (signal) Accuracy Adaptive systems Superresolution Bit rate Pipelines

来源：评论

学校读者我要写书评

暂无评论

Detection and classification of pneumonia using novel Superior Exponential (SupEx) activation function in convolutional neural networks

引用

EXPERT SYSTEMS WITH applications 2023年第1期217卷

作者： Kilicarslan, Serhat Kozkurt, Cemil Bas, Selcuk Elen, Abdullah Bandirma Onyedi Eylul Univ Dept Software Engn Fac Engn & Nat Sci Bandirma Balikesir Turkiye Bandirma Onyedi Eylul Univ Dept Transportat Engn Fac Engn & Nat Sci Bandirma Turkiye Bandirma Onyedi Eylul Univ Dept Accounting & Tax Bandirma Vocational Sch Bandirma Balikesir Turkiye

Many diseases threaten human health in many ways. Pneumonia is accepted as one of the most important health problems and causes of death worldwide. Therefore, deep learning techniques are used to solve problems in various fields, including diagnosing pneumonia. Convolutional neural networks (CNN), an artificial intelligence technique, are widely used in many areas such as segmentation, classification, signal processing. In deep networks, activation functions have an import. In CNN architectures, while the activation functions (AFs) are being developed, the activation functions are developed by taking into account the features such as not getting stuck in the local minimum of the CNN model and increasing the training performance. In the literature, although the ReLU is used in most studies, ReLU faces the problem that negative weights cannot be added to the network. To overcome the problem, the effect of the proposed new AFs on the real-world problem was investigated. In this study, a method for detecting pneumonia is suggested based on the novel proposed activation function and different CNN models. In the experimental studies, pneumonia was detected from chest X-ray images using ReLU, LReLU, Mish, Sigmoid, Swish, Smish, Logish, Softplus and the proposed activation function Superior Exponential (SupEx) were compared. Additionally, experimental study was performed through the MNIST and CIFAR-10 datasets to support the success of the proposed SupEx AF on the results. In the results obtained, it is seen that the CNN models used with the proposed SupEx have the best performance in both the detection of pneumonia from chest X-ray dataset and the MNIST dataset. The outstanding aspect of this study is that it is proposed to detect pneumonia with a new activation function, and it has been demonstrated that this activation function also works successfully in traditional benchmark datasets.

关键词： Convolutional neural networks (CNN) Activation function Pneumonia X-ray image

来源：评论

学校读者我要写书评

暂无评论

A Comprehensive Review of Emotion Recognition Using Multimedia Parameters: Challenges, Computational Intelligence and applications

引用

SN Computer Science 2025年第5期6卷 1-29页

作者： Shah, Aditya Kumar Kadyan, Virender Kane, Lalit School of Computer Science UPES Dehradun Bidholi India MIT World Peace University Pune India Verve Group Inc. Bangalore India

Advancement in human–computer interactions (HCI) demands optimized systems that depend on devising fast and reliable human emotion recognition algorithms. images, audio, sensory physiological signals, and body gestures are some of the identified principal modalities that characterize human emotions. However, the fusion of different modalities (homogeneous and heterogeneous) towards emotion recognition is challenging research where the space of diversities in different modalities are explored. Apart from dimensionality reduction schemes, fitting feature extraction methods, and fusion through models are some challenges which impact the performance of HCI systems. With the capability of processing unstructured data and real-time system designs, various neural networks have been reported to be the preferable choice for development of a multimodal emotion recognition system. This paper presents a comprehensive review on emotion recognition works based on image, audio, and physiological modalities. It begins with a discussion on various emotions and their interpretation with the said modalities. Further, emotion recognition works have been explored in terms of established and crafted emotion features. Later features extraction techniques that range from general machine learning based schemes to the recent deep neural networks are presented. Consequently, the performance of various emotion recognition schemes has been reviewed for individual and multi modalities-based systems. This performance review goes higher for the models based on various modality fusions, i.e., audio–video, audio-text, and body gestures. The fusion schemes mentioned here have been analyzed keeping in mind the efficacy of modality combinations towards each human emotion recognition. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2025.

关键词： Affective computing artificial intelligence Audio–video signals Body gesture Emotion recognition Human–computer interaction Physiological signals

来源：评论

学校读者我要写书评

暂无评论

Multidimensional Convolution Operation with Synthetic Frequency Dimensions in Photonics

引用

Physical Review Applied 2022年第3期18卷 034088-034088页

作者： Lingling Fan Zhexin Zhao Kai Wang Avik Dutt Jiahui Wang Siddharth Buddhiraju Casey C. Wojcik Shanhui Fan Department of Electrical Engineering Ginzton Laboratory Stanford University Stanford California 94305 USA Department of Mechanical Engineering and Institute for Physical Science and Technology University of Maryland College Park Maryland 20742 USA Department of Applied Physics Stanford University Stanford California 94305 USA

The convolution operation is widely used in signal and image processing and represents the most computationally intensive step in convolutional neural networks. We introduce a scheme to achieve arbitrary convolution kernels in the synthetic frequency dimension with a simple setup consisting of a ring resonator incorporating a phase and an amplitude modulator. This scheme can be used to perform multidimensional convolutions. We provide an analytic approach that determines the required modulation profile for any convolution kernel. Our work points to a direction of using optical computing to remove the computational bottleneck in traditional electronic circuits and may be useful in improving machine-learning hardware in artificial-intelligence applications.

关键词： Photonics artificial neural networks Convolutional neural networks Machine learning Optical computing

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：