检索结果-内蒙古大学图书馆

SWG: an architecture for sparse weight gradient computation

science China(Information sciences) 2024年第2期67卷 302-321页

作者： Weiwei WU Fengbin TU Xiangyu LI Shaojun WEI Shouyi YIN School of Integrated Circuits Tsinghua University Department of Electronic and Computer Engineering The Hong Kong University of Science and Technology

On-device training for deep neural networks(DNN) has become a trend due to various user preferences and scenarios. The DNN training process consists of three phases, feedforward(FF), backpropagation(BP), and weight gradient(WG) update. WG takes about one-third of the computation in the whole training process. Current training accelerators usually ignore the special computation property of WG and process it in a way similar to FF/BP. Besides, the extensive data sparsity existing in WG, which brings opportunities to save computation, is not well explored. Nevertheless, exploiting the optimization opportunities would meet three underutilization problems, which are caused by(1) the mismatch between WG data dimensions and hardware parallelism,(2) the full sparsity, i.e., the sparsity of feature map(Fmap),error map(Emap), and gradient, and(3) the workload imbalance resulting from irregular sparsity. In this paper, we propose a specific architecture for sparse weight gradient(SWG) computation. The architecture is designed based on hierarchical unrolling and sparsity-aware(HUSA) dataflow to exploit the optimization opportunities of the special computation property and full data sparsity. In HUSA dataflow, the data dimensions are unrolled hierarchically on the hardware architecture. A valid-data trace(VDT) mechanism is embedded in the dataflow to avoid the underutilization caused by the two-sided input sparsity. The gradient is unrolled in PE to alleviate the underutilization induced by output sparsity while maintaining the data reuse opportunities. Besides, we design an intra-and inter-column balancer(IIBLC) to dynamically tackle the workload imbalance problem resulting from the irregular sparsity. Experimental results show that with HUSA dataflow exploiting the full sparsity, SWG achieves a speedup of 12.23× over state-of-the-art gradient computation architecture, Train Ware. SWG helps to improve the energy efficiency of the state-of-the-art training accelerator LNPU from

关键词： CNN training gradient computation sparsity architecture

来源：评论

学校读者我要写书评

暂无评论

Deep learning and vision transformers-based framework for breast cancer and subtype identification

引用

Neural Computing and Applications 2025年第16期37卷 9311-9330页

作者： Jahan, Ishrat Chowdhury, Muhammad E. H. Vranic, Semir Al Saady, Rafif Mahmood Kabir, Saidul Pranto, Zahid Hasan Mim, Sabiha Jahan Nobi, Sadia Farhana Department of Computer Science and Engineering Shahjalal University of Science and Technology Sylhet Bangladesh Department of Electrical Engineering Qatar University 2713 Doha Qatar College of Medicine QU Health Qatar University Doha Qatar Department of Electrical and Electronic Engineering University of Dhaka Dhaka1000 Bangladesh

Breast cancer, marked by uncontrolled cell growth in breast tissue, is the most common cancer among women and a second-leading cause of cancer-related deaths. Among its types, ductal and lobular carcinomas are the most prevalent, with invasive ductal carcinoma accounting for about 70–80% of cases and invasive lobular carcinoma for about 10–15%. Accurate identification is crucial for effective treatment but can be time-consuming and prone to interobserver variability. AI can rapidly analyze pathological images, providing precise, cost-effective identification, thus reducing the pathologists’ workload. This study utilizes a deep learning framework for advanced, automatic breast cancer detection and subtype identification. The framework comprises three key components: detecting cancerous patches, identifying cancer subtypes (ductal and lobular carcinoma), and predicting patient-level outcomes from whole slide images (WSI). The validation process includes visualization using Score-CAM to highlight cancer-affected areas prominently. Datasets include 111 WSIs (85 malignant from the Warwick HER2 dataset and 26 benign from pathologists). For subtype detection, there are 57 ductal and 8 lobular carcinoma cases. A total of 28,428 annotated patches were reviewed by two expert pathologists. Four pre-trained models—DenseNet-201, MobileNetV2, an ensemble of these two, and a Vision Transformer-based model—were fine-tuned and tested on the patches. Patient-level results were predicted using a majority voting technique based on the percentage of each patch type in the WSI. The Vision Transformer-based model outperformed other models in patch classification, achieving an accuracy of 96.74% for cancerous patch detection and 89.78% for cancer subtype classification. For WSI-based cancer classification, the majority voting method attained an F1-score of 99.06 and 96.13% for WSI-based cancer subtype classification. The proposed deep learning-based framework for advanced breast cancer det

关键词： Diseases

来源：评论

学校读者我要写书评

暂无评论

Machine learning in crime prediction

引用

Journal of Ambient Intelligence and Humanized Computing 2023年第3期14卷 2887-2913页

作者： Jenga, Karabo Catal, Cagatay Kar, Gorkem Department of Computer Engineering Bahcesehir University Istanbul Turkey Department of Computer Science and Engineering Qatar University Doha Qatar

Predicting crimes before they occur can save lives and losses of property. With the help of machine learning, many researchers have studied predicting crimes extensively. In this paper, we evaluate state-of-the-art crime prediction techniques that are available in the last decade, discuss possible challenges, and provide a discussion about the future work that could be conducted in the field of crime prediction. Although many works aim to predict crimes, the datasets they used and methods that are applied are numerous. Using a Systematic Literature Review (SLR) methodology, we aim to collect and synthesize the required knowledge regarding machine learning-based crime prediction and help both law enforcement authorities and scientists to mitigate and prevent future crime occurrences. We focus primarily on 68 selected machine learning papers that predict crime. We formulate eight research questions and observe that the majority of the papers used a supervised machine learning approach, assuming that there is prior labeled data, and however in some cases, there is no labeled data in real-world scenarios. We have also discussed the main challenges found while conducting some of the studies by the researchers. We consider that this research paves the way for further research to help governments and countries fight crime and decrease this for better safety and security. © 2023, The Author(s).

关键词： Forecasting

来源：评论

学校读者我要写书评

暂无评论

Zero Trust in 5G Networks: Principles, Challenges, and Opportunities

Zero Trust in 5G Networks: Principles, Challenges, and Oppor...

引用

2024 Resilience Week, RWS 2024

作者： Lyu, Moyan Farooq, Junaid University of Michigan-Dearborn College of Engineering and Computer Science Department of Electrical & Computer Engineering DearbornMI48128 United States

ISBN: (纸本)9798350388985

The deployment of fifth-generation (5G) networks across various industry verticals is poised to transform communication and data exchange, promising unparalleled speed and capacity. However, the security concerns related to the widespread adoption of 5G, particularly in mission-critical sectors, present significant challenges. This article investigates the potential of a Zero Trust (ZT) security philosophy as a viable countermeasure to these concerns. It delves into the practicalities of implementing ZT principles within 5G networks, with a specific focus on harnessing AI/ML technologies for proactive security measures, dynamic policy adaptations, and advanced risk assessments. Further, the article underscores the importance of developing a tailored ZT maturity model for 5G networks. Furthermore, the paper outlines key future research directions aimed at improving the ZT maturity of 5G deployments, contributing to the safe and secure integration of 5G technology in various sectors. © 2024 IEEE.

关键词： 5G mobile communication systems

来源：评论

学校读者我要写书评

暂无评论

Geyser Inspired Algorithm:A New Geological-inspired Meta-heuristic for Real-parameter and Constrained engineering Optimization

引用

Journal of Bionic engineering 2024年第1期21卷 374-408页

作者： Mojtaba Ghasemi Mohsen Zare Amir Zahedi Mohammad-Amin Akbari Seyedali Mirjalili Laith Abualigah Department of Electronics and Electrical Engineering Shiraz University of TechnologyShirazIran Department of Electrical Engineering Faculty of EngineeringJahrom UniversityJahromFrasIran Department of Electrical and Computer Engineering Tarbiat Modares UniversityTehranIran Department of Electrical and Computer Engineering University ofCyprusNicosiaCyprus Centre for Artificial Intelligence Research and Optimisation Torrens University AustraliaBrisbaneQLD 4006Australia University Research and Innovation Center Obuda University1034 BudapestHungary Department of Electrical and Computer Engineering Lebanese American UniversityByblos 13-5053Lebanon Hourani Center for Applied Scientific Research Al-Ahliyya Amman UniversityAmman 19328Jordan MEU Research Unit Middle East UniversityAmman 11831Jordan Applied Science Research Center Applied Science Private UniversityAmman 11931Jordan

Over the past years,many efforts have been accomplished to achieve fast and accurate meta-heuristic algorithms to optimize a variety of real-world *** study presents a new optimization method based on an unusual geological phenomenon in nature,named Geyser inspired Algorithm(GEA).The mathematical modeling of this geological phenomenon is carried out to have a better understanding of the optimization *** efficiency and accuracy of GEA are verified using statistical examination and convergence rate comparison on numerous CEC 2005,CEC 2014,CEC 2017,and real-parameter benchmark ***,GEA has been applied to several real-parameter engineering optimization problems to evaluate its *** addition,to demonstrate the applicability and robustness of GEA,a comprehensive investigation is performed for a fair comparison with other standard optimization *** results demonstrate that GEA is noticeably prosperous in reaching the optimal solutions with a high convergence rate in comparison with other well-known nature-inspired algorithms,including ABC,BBO,PSO,and *** that the source code of the GEA is publicly available at https://***/projects/gea.

关键词： Nature-inspired algorithms Real-world and engineering optimization Mathematical modeling Geyser algorithm(GEA)

来源：评论

学校读者我要写书评

暂无评论

LRS: Enhancing Adversarial Transferability through Lipschitz Regularized Surrogate 38

LRS: Enhancing Adversarial Transferability through Lipschitz...

引用

38th AAAI Conference on Artificial Intelligence, AAAI 2024

作者： Wu, Tao Luo, Tie Wunsch, Donald C. Department of Computer Science Missouri University of Science and Technology United States Department of Electrical and Computer Engineering Missouri University of Science and Technology United States

ISBN: (纸本)1577358872

The transferability of adversarial examples is of central importance to transfer-based black-box adversarial attacks. Previous works for generating transferable adversarial examples focus on attacking given pretrained surrogate models while the connections between surrogate models and adversarial trasferability have been overlooked. In this paper, we propose Lipschitz Regularized Surrogate (LRS) for transfer-based black-box attacks, a novel approach that transforms surrogate models towards favorable adversarial transferability. Using such transformed surrogate models, any existing transfer-based black-box attack can run without any change, yet achieving much better performance. Specifically, we impose Lipschitz regularization on the loss landscape of surrogate models to enable a smoother and more controlled optimization process for generating more transferable adversarial examples. In addition, this paper also sheds light on the connection between the inner properties of surrogate models and adversarial transferability, where three factors are identified: smaller local Lipschitz constant, smoother loss landscape, and stronger adversarial robustness. We evaluate our proposed LRS approach by attacking state-of-the-art standard deep neural networks and defense models. The results demonstrate significant improvement on the attack success rates and transferability. Our code is available at https://***/TrustAIoT/LRS. Copyright © 2024, Association for the Advancement of Artificial Intelligence (***). All rights reserved.

关键词： Deep neural networks

来源：评论

学校读者我要写书评

暂无评论

When Crowdsourcing Meets Data Markets:A Fair Data Value Metric for Data Trading

引用

Journal of computer science & Technology 2024年第3期39卷 671-690页

作者：刘洋溯郑臻哲吴帆陈贵海 Department of Computer Science and Engineering Shanghai Jiao Tong UniversityShanghai 200240China CCF ACM IEEE

Large-quantity and high-quality data is critical to the success of machine learning in diverse *** with the dilemma of data silos where data is difficult to circulate,emerging data markets attempt to break the dilemma by facilitating data exchange on the ***,on the other hand,is one of the important methods to efficiently collect large amounts of data with high-value in data *** this paper,we investigate the joint problem of efficient data acquisition and fair budget distribution across the crowdsourcing and data *** propose a new metric of data value as the uncertainty reduction of a Bayesian machine learning model by integrating the data into model *** by this data value metric,we design a mechanism called Shapley Value Mechanism with Individual Rationality(SV-IR),in which we design a greedy algorithm with a constant approximation ratio to greedily select the most cost-efficient data brokers,and a fair compensation determination rule based on the Shapley value,respecting the individual rationality *** further propose a fair reward distribution method for the data holders with various effort levels under the charge of a data *** demonstrate the fairness of the compensation determination rule and reward distribution rule by evaluating our mechanisms on two real-world *** evaluation results also show that the selection algorithm in SV-IR could approach the optimal solution,and outperforms state-of-the-art methods.

关键词： data trading crowdsourcing mechanism design Shapley value

来源：评论

学校读者我要写书评

暂无评论

FAU Hack-a-Thons: An Introduction to Computational and Hardware, Logistical Skills, and Intense Training for Outcomes-Based Learning for Developing Internet of Things Products

FAU Hack-a-Thons: An Introduction to Computational and Hardw...

引用

2024 ASEE Annual Conference and Exposition

作者： Weinthal, Charles Perry Jaramillo, David Department of Computer & Electrical Engineering & Computer Science Florida Atlantic University Boca RatonFL United States

In preparation for the upcoming FAU Hack-a-Thon, we have implemented extensive support structures to ensure that all participating teams are thoroughly prepared for the competition. This preparation includes the provision of online support from a variety of programming experts from IBM and experienced Hack-a-Thon participants. This support is accessible 24/7 in the three weeks leading up to the event, aimed at ensuring all teams are up to speed and ready to participate. Our proactive approach focuses on encouraging teams to engage deeply with the training materials, consistently reminding them of their capability to successfully compete with adequate preparation. During the event, teams are expected to utilize the skills they have honed in areas such as teamwork, programming, and hardware development to create innovative Internet of Things (IoT) products. The support continues throughout the 24-hour event, with resources readily available to assist teams in overcoming any technical challenges they encounter. This setup aims to simulate real-world problem-solving scenarios that participants are likely to face in their professional careers, thus enhancing their experience and skills in a practical, hands-on manner. The Hack-a-Thon not only focuses on technical skills development but also emphasizes the cultivation of soft skills such as communication, teamwork, and time management. By the end of the event, participants are expected to present their projects to a panel of industry experts and faculty, demonstrating not only their technical prowess but also their ability to communicate their ideas effectively. This comprehensive approach ensures that participants leave the Hack-a-Thon with enhanced skills, increased confidence, and a significant competitive edge in the job market, as many past participants have noted the value of this experience during job interviews. © American Society for engineering Education, 2024.

关键词： Personnel training

来源：评论

学校读者我要写书评

暂无评论

Efficient video summarization through MobileNetSSD: a robust deep learning-based framework for efficient video summarization focused on objects of interest

引用

Multimedia Tools and Applications 2024年 1-26页

作者： Yarrarapu, Manasa Leelavathy, Narkedamilly Haritha, Dasari Department of Computer Science and Engineering Jawaharlal Nehru Technological University Andhra Pradesh Kakinada533003 India Department of Computer Science and Engineering Godavari Institute of Engineering and Technology Andhra Pradesh Rajahmundry533296 India Department of Computer Science and Engineering University College of Engineering Jawaharlal Nehru Technological University Andhra Pradesh Kakinada533003 India

Now-a-days, the generation of videos has increased dramatically due to the quick growth of multimedia and the internet. The need for effective ways to store, manage, and index the massive numbers of videos has become imperative due to this expansion. As a result, a method needs to be proposed that collects only the necessary data from the original recording. In computer vision, Video summarization is a significant task, and its primary goal is to give a quick summary of the video by removing irrelevant information and capturing key frames from the video. Many approaches have developed over the last several decades, using the most recent deep neural network architectures that represent the current state-of-the-art. Our method involves extracting vital key frames from the input video using the MobileNetSSD model, which is well-known for its efficient recognition and localization of objects of interest. These highlighted frames are essential in creating a detailed video summary. Furthermore, a method of temporal analysis is applied to guarantee that the summary accurately reflects the relevant events in the order in which they occurred, contributing to a coherent and meaningful representation of the information. We evaluated the proposed approach on TV Sum and SUM me video datasets, comparing the results against cutting-edge video summarization techniques. Our approach works effectively to produce clear and meaningful video summaries. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2024.

关键词： Network architecture

来源：评论

学校读者我要写书评

暂无评论

Optimizing Service Stipulation Uncertainty with Deep Reinforcement Learning for Internet Vehicle Systems

引用

computers, Materials & Continua 2023年第3期74卷 5705-5721页

作者： Zulqar Nain B.Shahana Shehzad Ashraf Chaudhry P.Viswanathan M.S.Mekala Sung Won Kim Department of Information and Communication Engineering Yeungnam UniversityGyeongsan-siKorea Department of Computer Science and Engineering KLEFIndia Department of Computer Engineering Faculty of Engineering and ArchitectureIstanbul Gelisim UniversityIstanbul34310Turkey School of Computer Science and Engineering VIT UniversityVelloreIndia

Fog computing brings computational services near the network edge to meet the latency constraints of cyber-physical System(CPS)*** devices enable limited computational capacity and energy availability that hamper end user *** designed a novel performance measurement index to gauge a device’s resource *** examination addresses the offloading mechanism issues,where the end user(EU)offloads a part of its workload to a nearby edge server(ES).Sometimes,the ES further offloads the workload to another ES or cloud server to achieve reliable performance because of limited resources(such as storage and computation).The manuscript aims to reduce the service offloading rate by selecting a potential device or server to accomplish a low average latency and service completion time to meet the deadline constraints of sub-divided *** this regard,an adaptive online status predictive model design is significant for prognosticating the asset requirement of arrived services to make float ***,the development of a reinforcement learning-based flexible x-scheduling(RFXS)approach resolves the service offloading issues,where x=service/resource for producing the low latency and high performance of the *** approach to the theoretical bound and computational complexity is derived by formulating the system efficiency.A quadratic restraint mechanism is employed to formulate the service optimization issue according to a set ofmeasurements,as well as the behavioural association rate and adulation *** system managed an average 0.89%of the service offloading rate,with 39 ms of delay over complex scenarios(using three servers with a 50%service arrival rate).The simulation outcomes confirm that the proposed scheme attained a low offloading uncertainty,and is suitable for simulating heterogeneous CPS frameworks.

关键词： Fog computing task allocation measurement models feasible node selection methods performance metrics

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：