检索结果-内蒙古大学图书馆

Annual Summit and Conference of the Asia-Pacific-signal-and-Information-processing-Association (APSIPA ASC)

作者： Huang, Yu-Min Tseng, Huan-Hsin Chien, Jen-Tzung Natl Chiao Tung Univ Dept Elect & Comp Engn Hsinchu Taiwan Univ Michigan Dept Radiat Oncol Ann Arbor MI 48109 USA

ISBN: (纸本)9781728132488

Spatial image and optical how provide complementary information for video representation and classification. Traditional methods separately encode two stream signals and then fuse them at the end of streams. This paper presents a new multi-stream recurrent neural network where streams are tightly coupled at each time step. Importantly, we propose a stochastic fusion mechanism for multiple streams of video data based on the Gumbel samples to increase the prediction power. A stochastic backpropagation algorithm is implemented to carry out a multi-stream neural network with stochastic fusion based on a joint optimization of convolutional encoder and recurrent decoder. Experiments on UCF101 dalaset illustrate the merits of the proposed stochastic fusion in recurrent neural network in terms of interpretation and classification performance.

关键词： Recurrent neural networks

来源：评论

学校读者我要写书评

暂无评论

No-reference image Quality Assessment Based on a Multi-feature Extraction Network 20

No-reference Image Quality Assessment Based on a Multi-featu...

引用

2nd International Conference on image, Video and signal processing, IVSP 2020

作者： Zhang, Hainan Meng, Fang Han, Yawen Communication University of China School of Information and Communication Engineering Beijing China Communication University of China School Information and Communication Engineering Beijing China

ISBN: (纸本)9781450376952

Deep convolutional neural network (DCNN) has achieved high performance on computer vision. However, it's hard to directly apply to image quality assessment due to lack of enough subjective scores. In this paper, we tackle this problem by exploiting high-level semantic and low-level structural features of the pre-trained VGG16 model for no-reference image quality assessment. We first divide overlapping 224x224 patches to get a fixed input size, then use VGG16 to extract the patch features. We use a three fully-connected layers network to get each patch quality score and average the patch scores to predict the whole image quality score. Experimental results on the benchmark LIVE-ii database show our methods are comparable with start-of-art algorithms and simpler than other DCNN feature-based methods. © 2020 ACM.

关键词： Convolution

来源：评论

学校读者我要写书评

暂无评论

Monolithic-3D Inference Engine with IGZO Based Ferroelectric Thin Film Transistor Synapses

TechRxiv

引用

TechRxiv 2022年

作者： De, Sourav Lederer, Maximilian Raffel, Yannick Lehninger, David Thunder, Sunanda Jank, Michael P.M. Ali, Tarek Kämpfe, Thomas Seidel, Konrad Sanctis, Shawn Fraunhofer-Institut für Photonische Mikrosysteme IPMS Center Nanoelectronic Technologies Dresden Germany Fraunhofer-Institut für Integrierte Systeme und Bauelementetechnologie Erlangen Germany

Instigated by the plethora of data generated by edge devices and IoT devices, machine learning has become the de facto choice of everyone for solving many tasks. Applications such as intelligent healthcare monitoring systems, smart watches, or autonomous cars require real-time processing of the data or image, which is done by machine learning algorithms with higher efficiency than humans. There are two possible methods for artificial intelligence-1. non-von-Neumann hardware-based implementation of neural networks 2. traditional computer science-based approach for neural networks or traditional von Neumann architecture-based implementation of neural networks The standard von-Neumann performance of neural networks, where the memory and computation parts are segregated, suffers from severe latency with an increasing number of edge devices. However, the multitude of edge devices used in our daily life imposes strict restrictions on latency, device area, and power consumption for hardware. Therefore, we need to take the route beyond the CMOS-based mixed-signal implementation of neural networks, where the memory bandwidth is not limited by the quintessential von Neumann architecture. The primary motivation of present-day research on the non-von-Neumann computing architecture is to build dedicated hardware modules for implementing low-power, fast-computing units without affecting the recent trend of scaling. This chapter focuses on amorphous indium-gallium-zinc-oxide (α-IGZO) based and their system-level applications. Back-end-of-line (BEoL) compatible indium IGZO-based multibit one-time programmable (OTP) ferroelectric thin film transistors (FeTFT) with lifelong retention capability were fabricated. The maximum temperature of the entire fabrication process was limited to 350oC. The gate-stack engineering by varying the thickness ratio of ferroelectric hafnium zirconium oxide (HZO) and IGZO layer fomented excellent data-retention capability and one-time programming propert

关键词： ii-VI semiconductors

来源：评论

学校读者我要写书评

暂无评论

Use of images Augmentation and Implementation of Doubly stochastic Models for Improving Accuracy of Recognition Algorithms Based on Convolutional neural Networks

Use of Images Augmentation and Implementation of Doubly Stoc...

引用

Systems of signal Synchronization, Generating and processing in Telecommunications (SYNCHROINFO)

作者： V. E. Dementyiev N. A. Andriyanov K. K. Vasilyiev Ulyanovsk State Technical University Ulyanovsk Russia JSC “RPC “Istok” named after Shokin” Fryazino Russia

ISBN: (数字)9781728160726

ISBN: (纸本)9781728160733

The article proposes methods for increasing the efficiency of pattern recognition in images using convolutional neural networks in conditions of insufficient reference images. It is proposed to use various kinds of image transformations to generate new images and increase the training sample volume. In particular, methods of scaling, adding noise, blurring, etc. are proposed. In addition to standard image transformations, it is supposed to increase the database by presenting patterns using a doubly stochastic model. In addition, the need for the use of regularization in training is noted. The recognition results are compared for two convolutional neural networks that differ only in the number of neurons for different initial data. It is shown that the highest recognition accuracy is provided by a network trained on a large number of augmentations in combination with images generated by a doubly stochastic model.

关键词： stochastic processes Mathematical model Training Neurons image recognition Biological neural networks Cats

来源：评论

学校读者我要写书评

暂无评论

Regularizing neural Networks by stochastically Training Layer Ensembles

Regularizing Neural Networks by Stochastically Training Laye...

引用

IEEE Workshop on Machine Learning for signal processing

作者： Alex Labach Shahrokh Valaee University of Toronto Toronto Canada

ISBN: (数字)9781728166629

ISBN: (纸本)9781728166636

Dropout and similar stochastic neural network regularization methods are often interpreted as implicitly averaging over a large ensemble of models. We propose STE (stochastically trained ensemble) layers, which enhance the averaging properties of such methods by training an ensemble of weight matrices with stochastic regularization while explicitly averaging outputs. This provides stronger regularization with no additional computational cost at test time. We show consistent improvement on various image classification tasks using standard network topologies.

关键词： Standards Training Biological neural networks Neurons Computational modeling stochastic processes Tuning

来源：评论

学校读者我要写书评

暂无评论

Vehicle logo recognition using whitening transformation and deep learning

引用

signal image AND VIDEO processing 2019年第1期13卷 111-119页

作者： Soon, Foo Chong Khaw, Hui Ying Chuah, Joon Huang Kanesan, Jeevan Univ Malaya Fac Engn Dept Elect Engn Kuala Lumpur 50603 Malaysia

This paper presents a vehicle logo recognition using a deep convolutional neural network (CNN) method and whitening transformation technique to remove redundancy of adjacent image pixels. Backpropagation algorithm with stochastic gradient descent optimization technique has been deployed to train and obtain weight filters of the networks. Seven layers of our proposed CNN incorporating an input layer, five hidden layers and an output layer have been implemented to capture rich and discriminative information of vehicle logo images. Functioning as the output layer of the network, the softmax classifier is utilized to handle multiple classes of vehicle logo image. For a given vehicle logo image, the network provides the probability for each vehicle manufacturer to which the given logo image belongs. Unlike most of the common traditional methods that employ handcrafted visual features, our proposed method is able to automatically learn and extract high-level features for the classification task. The extracted features are discriminative sufficiently to perform well in various imaging conditions and complex scenes. We validate our proposed method by utilizing a public vehicle logo image dataset, which comprises 10,000 and 1500 vehicle logo images for training and validation objective, respectively. Experimental results based on our proposed method outperform other existing methods in terms of the computational cost and overall classification accuracy of 99.13%.

关键词： Vehicle logo recognition Deep learning Convolutional neural network Optimization Whitening transformation

来源：评论

学校读者我要写书评

暂无评论

Improvement of edge-tracking methods using Genetic algorithm and neural network 5

Improvement of edge-tracking methods using Genetic algorithm...

引用

5th Iranian Conference on signal processing and Intelligent Systems (ICSPIS)

作者： Shabankareh, Sajjad Ghazanfari Shabankareh, Saeid Ghazanfari Islamic Azad Univ Fac Elect Engn Shiraz Iran Aix Marseille Univ Fac Sci & Technol Marseille France

ISBN: (纸本)9781728153506

One of the most basic and important operations in the field of image processing is image extraction and detection. Edge recognition is very important for image clarity and image segmentation. The importance of edge detection is that the human eye can recognizes the existence of different objects by observing the edges. So it makes sense to do edge detection before interpreting images in automated systems. Edges are points in the image where the two pixels are two different values, or two values very large in numerical value. It can be said that one of the goals of edge reduction in data size in images is to preserve the original structure and shape of the images. Editing has various applications, including object recognition, segmentation and image coding, in a variety of medical images. One of the problems we encounter when editing images has noises. In this paper, a combination of several standard edge- matching algorithms, neural network and multi-objective genetic algorithm Non-Dominated Sorting Genetic Algorithm (NSGA-ii) is used to edge detection. In the proposed method, we give the standard multi-edge finder to a forward propagating(FP) neural network as input. A multi-objective genetic algorithm with Non-Dominated Sorting Genetic Algorithm (NSGA-ii was used to select the number of edge detectors to access the neural network and to select the most accurate ones. The genetic algorithm selects the least accurate number of edge detectors to enter the neural network. To evaluate the proposed method, TP, TN, FP, FN criteria were used to compare with other methods. Finally, by comparing the results obtained on different images using existing methods and the proposed method it is observed that the proposed method has better accuracy in detecting the edge of noise images.

关键词： Edge detection neural network multi-objective genetic algorithm with Non-Dominated Sorting Genetic Algorithm (NSGA-ii) image processing

来源：评论

学校读者我要写书评

暂无评论

Deep Product Quantization Module for Efficient image Retrieval

Deep Product Quantization Module for Efficient Image Retriev...

引用

IEEE International Conference on Acoustics, Speech and signal processing

作者： Meihan Liu Yongxing Dai Yan Bai Ling-Yu Duan Institute of Digital Media Peking University China

ISBN: (数字)9781509066315

ISBN: (纸本)9781509066322

Product Quantization (PQ) is one of the most popular Approximate Nearest Neighbor (ANN) methods for large-scale image retrieval, bringing better performance than hashing based methods. In recent years, several works extend the hard quantization to soft quantization with specially designed deep neural architectures. We propose a simple but effective deep Product Quantization Module (PQM) to jointly learn discriminative codebook and precise hard assignment in an end-to-end manner. In this work, we use the straight-through estimator to make it feasible to directly optimize the discrete binary representations in deep neural networks with stochastic gradient descent. Different from previous deep vector quantization methods, PQM is a plug-and-play module which can be adaptive to various base networks in the scenarios of image search or compression. Besides, we propose a reconstruction loss to minimize the domain gap between the original embedding features and codebook. Experimental results show that PQM outperforms state-of-the-art deep supervised hashing and quantization methods on several image retrieval benchmarks.

关键词：

来源：评论

学校读者我要写书评

暂无评论

IMPROVING SUPER RESOLUTION methods VIA INCREMENTAL RESIDUAL LEARNING 26

IMPROVING SUPER RESOLUTION METHODS VIA INCREMENTAL RESIDUAL ...

引用

26th IEEE International Conference on image processing (ICIP)

作者： Aadil, Muneeb Rahim, Rafia Hussain, Sibt Ul Natl Univ Comp & Emerging Sci Reveal Ai Lab Islamabad Pakistan

ISBN: (纸本)9781538662496

Recently, Convolutional neural Networks (CNNs) have shown promising performance in super-resolution (SR). However, these methods operate primarily on Low Resolution (LR) inputs for memory efficiency but this limits, as we demonstrate, their ability to (i) model high frequency information;and (ii) smoothly translate from LR to High Resolution (HR) space. To this end, we propose a novel Incremental Residual Learning (IRL) framework to address these mentioned issues. In IRL, first we select a typical SR pre-trained network as a master branch. Next we sequentially train and add residual branches to the main branch, where each residual branch is learned to model accumulated residuals of all previous branches. We plug state of the art methods in IRL framework and demonstrate consistent performance improvement on public benchmark datasets to set a new state of the art for SR at only approximate to 20% increase in training time.(1)

关键词： image Reconstruction Convolutional neural Networks Residual Learning Super Resolution

来源：评论

学校读者我要写书评

暂无评论

SSGD: Sparsity-Promoting stochastic Gradient Descent Algorithm for Unbiased Dnn Pruning

SSGD: Sparsity-Promoting Stochastic Gradient Descent Algorit...

引用

IEEE International Conference on Acoustics, Speech and signal processing

作者： Ching-Hua Lee Igor Fedorov Bhaskar D. Rao Harinath Garudadri Department of ECE University of California San Diego ARM ML Research

ISBN: (数字)9781509066315

ISBN: (纸本)9781509066315;9781509066322

While deep neural networks (DNNs) have achieved state-of-the-art results in many fields, they are typically over-parameterized. Parameter redundancy, in turn, leads to inefficiency. Sparse signal recovery (SSR) techniques, on the other hand, find compact solutions to over-complete linear problems. Therefore, a logical step is to draw the connection between SSR and DNNs. In this paper, we explore the application of iterative reweighting methods popular in SSR to learning efficient DNNs. By efficient, we mean sparse networks that require less computation and storage than the original, dense network. We propose a reweighting framework to learn sparse connections within a given architecture without biasing the optimization process, by utilizing the affine scaling transformation strategy. The resulting algorithm, referred to as Sparsity-promoting stochastic Gradient Descent (SSGD), has simple gradient-based updates which can be easily implemented in existing deep learning libraries. We demonstrate the sparsification ability of SSGD on image classification tasks and show that it outperforms existing methods on the MNIST and CIFAR-10 datasets.

关键词： compressed sensing gradient methods learning (artificial intelligence) neural nets

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：