检索结果-内蒙古大学图书馆

48th IEEE International conference on Acoustics, Speech and Signal processing, ICASSP 2023

作者： Yang, Jinhai Yang, Hua Shanghai Jiao Tong University Institution of Image Communication and Network Engineering China Shanghai Key Lab of Digital Media Processing and Transmission Shanghai China

ISBN: (纸本)9781728163277

Few-shot learning aims for rapid adaptation with few samples. Recently, cross-granularity few-shot learning has emerged as a promising research area, where models observe coarse labels but target fine-grained recognition among novel classes. As coarse supervision tends to eliminate feature discrimination among underlying sub-classes, existing methods commonly utilize self-supervision as a complement to explore intra-class variation. However, current methods suffer from an intrinsic conflict between contrastive learning and coarse supervision. In this paper, we locate the root cause of the intrinsic conflict. Then, we resolve it by exploiting the similarity among augmented views while ignoring the unreasonable constraint between negative pairs. Besides, we decouple contrastive learning and coarse supervision into parallel branches to better regularize the latent space. Albeit simple, our approach consistently outperforms state-of-the-art methods across different benchmarks. © 2023 IEEE.

关键词： cross-granularity transfer Few-shot learning generalization image classification

来源：评论

学校读者我要写书评

暂无评论

Invisible Backdoor Attack with image Contours Triggers 24th

Invisible Backdoor Attack with Image Contours Triggers

引用

24th International conference on Algorithms and Architectures for parallel processing, ICA3PP 2024

作者： Hou, Yangzhi Yang, Ying Jinan China Shandong Provincial Key Laboratory of Computing Power Internet and Service Computing Shandong Fundamental Research Center for Computer Science Jinan China

ISBN: (纸本)9789819615445

In recent years, backdoor attack techniques on neural networks have been widely studied and researched. In this attack mode, the model implanted with a backdoor behaves normally when processing normal inputs, but once it encounters a special trigger designed by the attacker, its prediction results will be maliciously manipulated. Different backdoor attack strategies have been proposed to improve the stealth and robustness of the triggers. However, most of the existing backdoor attack strategies use fixed triggering patterns or input-independent triggering patterns, and thus are easily detected and blocked by existing defense systems. To address these limitations, we propose a backdoor attack method based on image contours. Specifically, we first use image contours as the target region for backdoor injection and fill them with specific color information to generate trigger patterns. Then, using our designed Generative Network (GN) and the Auxiliary Extraction Network (AE), the trigger patterns are covertly embedded into the images to be contaminated to generate a poisoned dataset, which contaminates the training model for the purpose of malicious attacks. Compared with traditional methods, our trigger patterns are more natural and harder to detect. Experimental results show that our proposed attack method achieves excellent performance in terms of both stealth and robustness. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

Frequency Decoupled Domain-Irrelevant Feature Learning for Pan-Sharpening

引用

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY 2025年第2期35卷 1237-1250页

作者： Zhang, Jie Cao, Ke Yan, Keyu Lin, Yunlong He, Xuanhua Wang, Yingying Li, Rui Xie, Chengjun Zhang, Jun Zhou, Man Chinese Acad Sci Inst Intelligent Machines Hefei Inst Phys Sci Hefei 230031 Peoples R China Univ Sci & Technol China Grad Sch Sci Isl Branch Hefei 230026 Peoples R China Chinese Acad Sci Inst Intelligent Machines Hefei Inst Phys Sci Hefei 230031 Peoples R China Xiamen Univ Sch Informat Xiamen 361005 Peoples R China Xiamen Univ Key Lab Multimedia Trusted Percept & Efficient Com Minist Educ China Xiamen 361005 Peoples R China Chinese Acad Sci High Magnet Field Lab Hefei Inst Phys Sci Hefei 230031 Peoples R China Univ Sci & Technol China Dept Automat Hefei 230026 Peoples R China

Pan-sharpening aims to generate high-detail multi-spectral images (HRMS) through the fusion of panchromatic (PAN) and multi-spectral (MS) images. However, existing pan-sharpening methods often suffer from significant performance degradation when dealing with out-of-distribution data, as they assume the training and test datasets are independent and identically distributed. To overcome this challenge, we propose a novel frequency domain-irrelevant feature learning framework that exhibits exceptional generalization capabilities. Our approach involves parallel extraction and processing of domain-irrelevant information from the amplitude and phase components of the input images. Specifically, we design a frequency information separation module to extract the amplitude and phase components of the paired images. The learnable high-pass filter is then employed to eliminate domain-specific information from the amplitude spectrums. After that, we devised two specialized sub-networks (AFL-Net and PFL-Net) to perform targeted learning of the frequency domain-irrelevant information. This allows our method to effectively capture the complementary domain-irrelevant information contained in the amplitude and phase spectra of the images. Finally, the information fusion and restoration module dynamically adjusts the feature channel weights, enabling the network to output high-quality HRMS images. Through this frequency domain-irrelevant feature learning framework, our method balances generalization capability and network performance on the distribution of training dataset. Extensive experiments conducted on various satellite datasets demonstrate the effectiveness of our method for generalized pan-sharpening. Our proposed network outperforms state-of-the-art methods in terms of both quantitative metrics and visual quality, showcasing its superior ability to handle diverse, out-of-distribution data.

关键词： Representation learning Frequency-domain analysis Feature extraction Data mining Training Spatial resolution Circuits and systems image restoration Fourier transforms Distortion Frequency domain-irrelevant pan-sharpening learnable filter amplitude spectrums

来源：评论

学校读者我要写书评

暂无评论

parallel Cholesky Factorization for Banded Matrices Using OpenMP Tasks 29th

Parallel Cholesky Factorization for Banded Matrices Using Op...

引用

29th International conference on parallel and distributed Computing (Euro-Par)

作者： Liu, Felix Fredriksson, Albin Markidis, Stefano KTH Royal Inst Technol Stockholm Sweden RaySearch Labs Stockholm Sweden

ISBN: (纸本)9783031396977;9783031396984

Cholesky factorization is a method for solving linear systems involving symmetric, positive-definite matrices, and can be an attractive choice in applications where a high degree of numerical stability is needed. One such application is mathematical optimization, where direct methods for solving linear systems are widely used and often a significant performance bottleneck. An example where this is the case, and the specific type of optimization problem motivating this work, is radiation therapy treatment planning, where mathematical optimization is used to create individual treatment plans for patients. To address this bottleneck, we propose a task-based multi-threaded method for Cholesky factorization of banded matrices with medium-sized bands. We implement our algorithm using OpenMP tasks and compare our performance with state-of-the-art libraries such as Intel MKL. Our performance measurements show a performance that is on par or better than Intel MKL (up to similar to 26% on a single CPU socket) for a wide range of matrix bandwidths on two different Intel CPU systems.

关键词： Cholesky factorization Task-Based parallelism Linear Solver

来源：评论

学校读者我要写书评

暂无评论

Research on evaluation method of image style transfer algorithm 3

Research on evaluation method of image style transfer algori...

引用

3rd International conference on Information Science, parallel and distributed Systems, ISPDS 2022

作者： Su, Na Pan, Yun School of Computer and Cyberspace Security Communication University of China Beijing China

ISBN: (数字)9781665487474

ISBN: (纸本)9781665487474

The task of style transfer is to transfer the style of the target image to the required image. In recent years, there have been increasing studies on image style transfer. and now it is often used in animation production, software interface beautification, image expansion and the creation of a variety of styles of decorative patterns. As these algorithms are proposed, how to evaluate algorithms will become an important problem, only with evaluation can comparative analysis be carried out and algorithms can continue to develop and progress. The importance of evaluation can not be underestimated. This paper firstly systematically combs and introduces the evaluation method of algorithm from subjective and objective aspects. Subjective methods are divided into two parts: questionnaire and User Study, which mainly rely on people's subjective perception and are easily influenced by subjects. The objective method evaluates the algorithm from a more accurate perspective, mainly including relying on statistics and evaluation indicators. Secondly, this paper introduces the experiments using several evaluation methods and evaluates their effects. Finally, it points out the problems existing in the current evaluation methods to provide the direction for the follow-up research. © 2022 IEEE.

关键词： image quality Information science Software algorithms Production Animation Software Task analysis

来源：评论

学校读者我要写书评

暂无评论

LCIP: A Retargetable Framework for Optimized CNN Inference

LCIP: A Retargetable Framework for Optimized CNN Inference

引用

conference on Real-Time image processing and Deep Learning

作者： Pan, Lei Zhang, Yan Bhattacharyya, Shuvra S. Univ Maryland College Pk MD 20742 USA

ISBN: (数字)9781510661714

ISBN: (纸本)9781510661707;9781510661714

Deep Neural Networks (DNNs) are adopted in numerous application areas of signal and information processing with Convolutional Neural Networks (CNNs) being a particularly popular class of DNNs. Many machine learning (ML) frameworks have evolved for design and training of CNN models, and similarly, a wide variety of target platforms, ranging from mobile and resource-constrained platforms to desktop and more powerful platforms, are used to deploy CNN-equipped applications. To help designers navigate the complex design spaces involved in deploying CNN models derived from ML frameworks on alternative processing platforms, retargetable methods for implementing CNN models are of increasing interest. In this paper, we present a novel software tool, called the Lightweight-dataflow-based CNN Inference Package (LCIP), for retargetable, optimized CNN inference on different hardware platforms (e.g., x86 and ARM CPUs, and GPUs). In LCIP, source code for CNN operators (convolution, pooling, etc.) derived from ML frameworks is wrapped within dataflow actors. The resulting coarse grain dataflow models are then optimized using the retargetable LCIP runtime engine, which employs higherlevel dataflow analysis and orchestration that is complementary to the intra-operator performance optimizations provided by the ML framework and the back-end development tools of the target platform. Additionally, LCIP enables heterogeneous and distributed edge inference of CNNs by offloading part of the CNN to additional devices, such as onboard GPU or network devices. Our experimental results show that LCIP provides significant improvements in inference throughput on commonly-used CNN architectures, and the improvement is consistent across desktop and resource-constrained platforms.

关键词： Convolutional Neural Networks distributed Inference Dataflow Modeling Edge Computing Heterogeneous Computing

来源：评论

学校读者我要写书评

暂无评论

Using Programmable P4 Switches to Reduce Communication Costs of parallel and distributed Simulations

Using Programmable P4 Switches to Reduce Communication Costs...

引用

IEEE Global Communications conference (GLOBECOM)

作者： Wang, Shie-Yuan Kuo, Nai-En Natl Yang Ming Chiao Tung Univ Dept Comp Sci Hsinchu Taiwan

ISBN: (纸本)9781665435406

In the past, several time synchronization methods have been proposed for parallel and distributed simulation (PDS). Among them, one widely used conservative method is the Chandy-Misra-Bryant (CMB) algorithm. In the CMB algorithm, many null messages may be exchanged among logical processes to advance their clocks so that deadlock will not occur among them. In this work, using a data-plane programmable P4 hardware switch, we design and implement a data fusion-based approach inside the packet processing pipeline of the P4 switch. Our approach extracts the timestamp carried in exchanged null messages, computes the fusion results of these timestamps, drops unnecessary null messages inside the switch, generates new messages carrying the fusion results, and sends these generated messages to only the logical processes that can benefit from receiving these messages. Experimental results show that on an 8-host testbed, our approach can speed up a PDS by a factor of 2.75 and 1.65 when compared with the unicast and multicast approaches, respectively.

关键词： software-defined network P4 data-plane programmable switch parallel and distributed simulation

来源：评论

学校读者我要写书评

暂无评论

An unsupervised deep learning framework for gene regulatory network inference from single-cell expression data

An unsupervised deep learning framework for gene regulatory ...

引用

2023 IEEE International conference on Bioinformatics and Biomedicine, BIBM 2023

作者： Mao, Guo Liu, Jie National University of Defense Technology Science and Technology on Parallel and Distributed Processing Laboratory Changsha410073 China National University of Defense Technology Science and Technology on Parallel and Distributed Processing Laboratory National University of Defense Technology Laboratory of Software Engineering for Complex System Changsha410073 China

ISBN: (纸本)9798350337488

Recent advances in single-cell RNA sequencing (scRNA-seq) technology provides unprecedented opportunities for reconstruction gene regulation networks (GRNs). At present, many different models have been proposed to infer GRN from a large number of RNA-seq data, but most deep learning models use a priori gene regulatory network to infer potential GRNs. It is a challenge to reconstruct GRNs from scRNA-seq data due to the noise and sparsity introduced by the dropout effect. Here, we propose GAALink, a novel unsupervised deep learning method. It first constructs the gene similarity matrix and then refines it by threshold value. It then learns feature representations of genes through a graphical attention autoencoder that propagates information across genes with different weights. Finally, we use gene feature expression for matrix completion such that the GRNs are reconstructed. Compared with seven existing GRNs reconstruction methods, GAALink achieves more accurate performance on seven scRNA-seq dataset with four ground truth networks. GAALink can provide a useful tool for inferring GRNs for scRNA-seq expression data. © 2023 IEEE.

关键词： Feature representations Gene similarity matrix Graphical attention autoencoder Matrix completion Unsupervised deep learning method

来源：评论

学校读者我要写书评

暂无评论

Large Scale image Classification Using GPU-based Genetic Programming 22

Large Scale Image Classification Using GPU-based Genetic Pro...

引用

Genetic and Evolutionary Computation conference (GECCO)

作者： Zeng, Peng Lensen, Andrew Sun, Yanan Sichuan Univ Coll Comp Sci Chengdu Peoples R China Victoria Univ Wellington Sch Engn & Comp Sci Wellington New Zealand

ISBN: (纸本)9781450392686

Genetic programming (GP) has been applied to image classification and achieved promising results. However, most GP-based image classification methods are only applied to small-scale image datasets because of the limits of high computation cost. Efficient acceleration technology is needed when extending GP-based image classification methods to large-scale datasets. Considering that fitness evaluation is the most time-consuming phase of the GP evolution process and is a highly parallelized process, this paper proposes a CPU multi-processing and GPU parallel approach to perform the process, and thus effectively accelerate GP for image classification. Through various experiments, the results show that the highly parallelized approach can significantly accelerate GP-based image classification without performance degradation. The training time of GP-based image classification method is reduced from several weeks to tens of hours, enabling it to be run on large-scale image datasets.

关键词： genetic programming image classification parallel algorithms

来源：评论

学校读者我要写书评

暂无评论

Field-VIO: Stereo Visual-Inertial Odometry Based on Quantitative Windows in Agricultural Open Fields

Field-VIO: Stereo Visual-Inertial Odometry Based on Quantita...

引用

IEEE International conference on Robotics and Automation (ICRA)

作者： Sun, Jianjing Wu, Shuang Dong, Jun He, Junming Anhui Univ Inst Phys Sci Hefei 230601 Peoples R China Chinese Acad Sci Hefei Inst Phys Sci Hefei 230031 Peoples R China

ISBN: (纸本)9798350384581;9798350384574

In agricultural open fields, accurate autonomous localization of robots requires long-term data correlation to reduce cumulative error. Our article presents a Stereo Visual-Inertial Odometry (VIO) system based on ORB-SLAM3 to address the malfunction of the Loop Closure Detection (LCD) methods in this environment. In this method, we first propose a concept of quantitative windows to describe the robot's trajectory along the crop rows. We design a driving state quantification algorithm and accurately separate the quantitative windows between the crop rows. Our system constructs spatial constraints according to the parallelism between the quantitative windows. We apply an anomaly correction method to maintain the constructed parallel matching relationship and implement holistic pose correction for keyframes within abnormal quantitative windows. Our system demonstrated excellent performance over long distances in experiments on the Rosario dataset, verifying its effectiveness in reducing cumulative positioning error in agricultural open fields.

关键词： Stereo 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：