检索结果-内蒙古大学图书馆

6th Asia Energy and Electrical Engineering Symposium (AEEES)

作者： Li, Bin Zhao, Ruifeng Qiu, Junqi Guangdong Power Grid Co Ltd Zhongshan Power Supply Bur Zhongshan Peoples R China Guangdong Power Grid Co Ltd Power Dispatch & Control Ctr Guangzhou Peoples R China

ISBN: (纸本)9798350373486;9798350373479

With the growing capacity of distributed renewable energy resources (RERs) integrated into distribution grids, the power flow distribution becomes more complex. Traditional fault diagnosis solutions are proving inadequate for the demands of the evolving distribution network, thereby diminishing the reliability and sensitivity of the protection system. To address this challenge, this paper introduces an innovative fault diagnosis approach for distribution networks incorporating RERs, leveraging signal processing techniques and machine learning algorithms. Initially, effective features are obtained from the measured current signals by the Hilbert-Huang transform (HHT). Subsequently, these fault features serve as inputs for training feed-forward neural networks to build fault diagnosis models (including detection, classification, and segment identification). Simulation tests are conducted on a 13-node distribution network with three different types of RERs. Simulation results show that the method can accurately diagnose distribution network faults, and is robust to fault inception angle variations, transition resistance, and noise interference.

关键词： distribution network fault diagnosis Hilbert-Huang transform machine learning renewable energy resource

来源：评论

学校读者我要写书评

暂无评论

Optimal Gradient Sliding and its Application to distributed Optimization Under Similarity 36

Optimal Gradient Sliding and its Application to Distributed ...

引用

36th Conference on neural Information processing Systems (NeurIPS)

作者： Kovalev, Dmitry Beznosikov, Aleksandr Borodich, Ekaterina Gasnikov, Alexander Scutari, Gesualdo King Abdullah Univ Sci & Technol Thuwal Saudi Arabia MIPT Moscow Russia HSE Univ Yandex Russia Inst Informat Transmiss Problems RAS Moscow Russia Purdue Univ W Lafayette IN 47907 USA

ISBN: (纸本)9781713871088

We study structured convex optimization problems, with additive objective r := p + q, where r is (mu-strongly) convex, q is L-q-smooth and convex, and p is L-p-smooth, possibly nonconvex. For such a class of problems, we proposed an inexact accelerated gradient sliding method that can skip the gradient computation for one of these components while still achieving optimal complexity of gradient calls of p and q, that is, O(root L-p/mu) and O(root L-q/mu), respectively. This result is much sharper than the classic black-box complexityO(root(L-p + L-q)/mu), especially when the difference between L-p and L-q is large. We then apply the proposed method to solve distributed optimization problems over master-worker architectures, under agents' function similarity, due to statistical data similarity or otherwise. The distributed algorithm achieves for the first time lower complexity bounds on both communication and local gradient calls, with the former having being a longstanding open problem. Finally the method is extended to distributed saddleproblems (under function similarity) by means of solving a class of variational inequalities, achieving lower communication and computation complexity bounds.

关键词： Convex optimization

来源：评论

学校读者我要写书评

暂无评论

Convolutional Recurrent neural networks for the Classification of Cetacean Bioacoustic Patterns 48

Convolutional Recurrent Neural Networks for the Classificati...

引用

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

作者： Makropoulos, Dimitris N. Tsiami, Antigoni Prospathopoulos, Aristides Kassis, Dimitris Frantzis, Alexandros Skarsoulis, Emmanuel Piperakis, George Maragos, Petros National Technical University of Athens Greece Anavyssos Greece Pelagos Cetacean Research Institute Vouliagmeni 16671 Greece IACM Foundation of Research and Technology - Hellas Crete Heraklion Greece

ISBN: (纸本)9781728163277

In this paper we focus on the development of a convolutional recurrent neural network (CRNN) to categorize biosignals collected in the Hellenic Trench, generated by two cetacean species, sperm whales (Physeter macrocephalus) and striped dolphins (Stenella coeruleoalba). We convert audio signals into mel-spectrograms and forward the input into a deep residual network (ResNet), designed to capture spectral patterns. Next, ResNet's output is reshaped into a time-distributed layer and fed into recurrent network variants, Long Short-Term Memory (LSTMs) or Gated Recurrent Units (GRUs), able to recognize long-term time dependencies on extracted features. The hybrid network perfectly classifies audio signals into three categories (dolphins, sperm whales, ambient noise) while it also exhibits high learning ability on recognising intraclass representations of overlapping acoustic patterns (clicks vs whistles and clicks, both emitted by dolphins). The proposed scheme outperforms traditional Machine Learning (ML) techniques, baseline ResNet and LSTM architectures or their deep parallel combinations. © 2023 IEEE.

关键词： Long short-term memory

来源：评论

学校读者我要写书评

暂无评论

HSDP: Accelerating Large-scale Model Training via Efficient Sharded Data Parallelism 22

HSDP: Accelerating Large-scale Model Training via Efficient ...

引用

22nd IEEE International Symposium on Parallel and distributed processing with Applications, ISPA 2024

作者： Hao, Yanqi Lai, Zhiquan Wang, Wei Li, Shengwei Liu, Weijie Ge, Keshi Li, Dongsheng College of Computer Changsha China

ISBN: (纸本)9798331509712

Large deep neural network (DNN) models have demonstrated exceptional performance across diverse downstream tasks. Sharded data parallelism (SDP) has been widely used to reduce the memory footprint of model states. In a DNN training cluster, a device usually has multiple inter-device links that connect to other devices, like NVLink and InfiniBand. However, existing SDP approaches employ a single link at any given time, encountering challenges in efficient training due to significant communication overheads. We observe that the inter-device links can work independently without affecting each other. To reduce the fatal communication overhead of distributed training of large DNNs, this paper introduces HSDP, an efficient SDP training approach that enables the simultaneous utilization of multiple inter-device links. HSDP partitions models in a novel fine-grained manner and orchestrates the communication processes of partitioned parameters while considering inter-device links. This design enables concurrent communication execution and reduces communication overhead. To further optimize the training performance of HSDP, we propose a HSDP planner. The HSDP planner first abstracts the model partition and execution of HSDP into a communication parallel strategy, and builds a cost model to estimate the performance of each strategy. We then formulate the strategy searching as an optimization problem and solve it with an off-the-shelf solver. Evaluations on representative DNN workloads demonstrate that HSDP achieves up to 1.30× speedup compared to the state-of-the-art SDP training approaches. © 2024 IEEE.

关键词： Deep neural networks

来源：评论

学校读者我要写书评

暂无评论

Advances in Thermal Imaging: A Convolutional neural network Approach for Improved Breast Cancer Diagnosis 2

Advances in Thermal Imaging: A Convolutional Neural Network ...

引用

2nd International Conference on distributed Computing and Optimization Techniques, ICDCOT 2024

作者： Agughasi, Victor Ikechukwu Bhimshetty, Sampoorna Deepu, R. Mala, M.V. Maharaja Institute of Technology Mysore Department of Computer Science & Engineering Karnataka India University of Visvesvaraya College of Engineering Department of Computer Science & Engineering Karnataka India Pes University Department of Computer Science & Engineering Karnataka Bangalore India

ISBN: (纸本)9798350382952

This study explores the application of thermal imaging in breast cancer diagnostics, presenting a novel methodology that integrates pre-processing techniques and Convolutional neural networks (CNN) for the classification of thermal breast images. A comprehensive dataset from the DMR Database is used, comprising balanced samples of normal and cancerous images. The images are augmented, standardized, and enhanced through data augmentation, resizing, and filtering, with crucial features extracted via the Histogram of Oriented Gradients (HOG). A CNN model, specifically constructed for this study, is then trained and tested on these processed images. The approach's effectiveness is benchmarked against other classifiers, displaying promising results with accuracy rates ranging from 95.7% to 98.5%. Future work suggests exploring the fusion of traditional and advanced imaging modalities, expanding datasets, and utilizing pre-trained models for improved diagnostic precision. This research signifies a step toward the development of real-time, efficient diagnostic tools for breast cancer, highlighting the potential impact on patient outcomes and the broader medical field. © 2024 IEEE.

关键词： Infrared imaging

来源：评论

学校读者我要写书评

暂无评论

Dance of SNN and ANN: Solving binding problem by combining spike timing and reconstructive attention 36

Dance of SNN and ANN: Solving binding problem by combining s...

引用

36th Conference on neural Information processing Systems (NeurIPS)

作者： Zheng, Hao Lin, Hui Zhao, Rong Shi, Luping Tsinghua Univ Dept Precis Instrument Ctr Brain Inspired Comp Res Beijing 100084 Peoples R China

ISBN: (纸本)9781713871088

The binding problem is one of the fundamental challenges that prevent the artificial neural network (ANNs) from a compositional understanding of the world like human perception, because disentangled and distributed representations of generative factors can interfere and lead to ambiguity when complex data with multiple objects are presented. In this paper, we propose a brain-inspired hybrid neural network (HNN) that introduces temporal binding theory originated from neuroscience into ANNs by integrating spike timing dynamics (via spiking neural networks, SNNs) with reconstructive attention (by ANNs). Spike timing provides an additional dimension for grouping, while reconstructive feedback coordinates the spikes into temporal coherent states. Through iterative interaction of ANN and SNN, the model continuously binds multiple objects at alternative synchronous firing times in the SNN coding space. The effectiveness of the model is evaluated on synthetic datasets of binary images. By visualization and analysis, we demonstrate that the binding is explainable, soft, flexible, and hierarchical. Notably, the model is trained on single object datasets without explicit supervision on grouping, but successfully binds multiple objects on test datasets, showing its compositional generalization capability. Further results show its binding ability in dynamic situations.

关键词： neural networks

来源：评论

学校读者我要写书评

暂无评论

SUSTAINABLE DEVELOPMENT IN MEDICAL APPLICATIONS USING neural network ARCHITECTURE

引用

SCALABLE COMPUTING-PRACTICE AND EXPERIENCE 2024年第2期25卷 782-791页

作者： Jiang, Shuyi Changchun Univ Sci & Technol Coll Comp Sci & Technol Changchun 130012 Peoples R China

The purpose of this research is to propose a methodology utilizing machine learning techniques to support medical organizations in effectively managing risks. Specifically, the study aims to connect social media data to identify and assess potential threats, ultimately enabling healthcare management to make informed decisions for their organizations and clients. The research employs machine learning algorithms to analyze user-generated content on social media platforms, generating comprehensive visual representations of various risk categories and their magnitudes. Additionally, the study utilizes data simplification techniques, including categorization, to streamline data processing and enhance overall efficiency. A computational framework is also developed, incorporating closed-form connections for threat assessment and evaluation. The study further empirically analyses the Consumer Value Stores (CVS) established for medical care in the United States. The findings reveal that prevalent threats within the lower quartile of client messages about CVS services include operational, financial, and technological risks. The severity of these risks is distributed among high risk (21.8%), moderate risk (78%), and minimal risk (0.2%). The research also presents several metrics to demonstrate the robustness of the proposed framework, confirming its effectiveness in effectively identifying and addressing potential threats. This research provides insights that can help healthcare management make informed decisions and foster a safer and more secure environment for their organizations and the people they serve.

关键词： Conversational Interpretation Artificial Intelligence Internet-based Information Mining Medical Analytics Risk Administration

来源：评论

学校读者我要写书评

暂无评论

On-the-fly (D)DoS attack mitigation in SDN using Deep neural network-based rate

引用

COMPUTER COMMUNICATIONS 2022年 182卷 153-169页

作者： El Kamel, Ali Eltaief, Hamdi Youssef, Habib Univ Sousse PRINCE Res Lab ISITC Hammam Sousse Sousse Tunisia

Software Defined networking (SDN) has emerged as a promising paradigm offering an unprecedented programmability, scalability and fine-grained control over forwarding elements (FE). Mainly, SDN decouples the forwarding plane from the control plane which is moved to a central controller that is in charge of taking routing decisions in the network. However, SDN is rife with vulnerabilities so that several network attacks, especially distributed Denial of Service (DDoS), can be launched from compromised hosts connected to switches. DDoS attacks can easily overload the controller processing capacity and flood switch flow-tables. This paper deals with the security issue in SDN. It proposes a real-time protection against DDoS attacks that is based on a controller-side sliding window rate limiting approach which relies on a weighted abstraction of the underlying network. A weight defines the allowable amount of data that can be transmitted by a node and is dynamically updated according to its contribution to: (1) the queueing capacity of the controller, and (2) the number of flow-rules in the switch. Hence, a new deep learning algorithm, denoted the Parallel Online Deep Learning algorithm (PODL), is defined in order to update weights on the-fly according to both aforementioned constraints simultaneously. Furthermore, the behavior of each host and each switch is evaluated through a measure of trustworthiness which is used to penalize mis-behaving ones by prohibiting new flow requests or PacketIn messages for a period of time. Host trustworthiness is based on their weights while switch trustworthiness is achieved through a computation of the Average Nearest-Neighbor Degree (ANND). Realistic experiments show that the proposed solution succeeds in minimizing the impact of DDoS attacks on both the controllers and the switches regarding the PacketIn arrival rate at the controller, the rate of accepted requests and the flow-table usage.

关键词： Software Defined networking (SDN) Deep Shallow neural networks (SNN) MultiLoss SNN (ML-SNN) Online Deep Learning (ODL) Rate limiting Trustworthiness

来源：评论

学校读者我要写书评

暂无评论

Matcha: A Matching-Based Link Scheduling Strategy to Speed up distributed Optimization

引用

IEEE TRANSACTIONS ON SIGNAL processing 2022年 70卷 5208-5221页

作者： Wang, Jianyu Sahu, Anit Kumar Joshi, Gauri Kar, Soummya Carnegie Mellon Univ Dept Elect & Comp Engn Pittsburgh PA 15213 USA Meta Platforms Menlo Pk CA 94025 USA Amazon Alexa AI Pittsburgh PA 15232 USA

In this paper, we study the problem of distributed optimization using an arbitrary network of lightweight computing nodes, where each node can only send/receive information to/from its direct neighbors. Decentralized stochastic gradient descent (SGD) has been shown to be an effective method to train machine learning models in this setting. Although decentralized SGD has been extensively studied, most prior works focus on the error-versus-iterations convergence, without taking into account how the topology affects the communication delay per iteration. For example, a denser (sparser) network topology results in faster (slower) error convergence in terms of iterations, but it incurs more (less) communication time per iteration. We propose MATCHA, an algorithm that can achieve a win-win in this error-runtime trade-off for any arbitrary network topology. The main idea of MATCHA is to communicate more frequently over connectivity-critical links in order to ensure fast convergence, and at the same time minimize the communication delay per iteration by using other links less frequently. It strikes this balance by decomposing the topology into matchings and then optimizing the set of matchings that are activated in each iteration. Experiments on a suite of datasets and deep neural networks validate the theoretical analyses and demonstrate that MATCHA takes up to 5x less time than vanilla decentralized SGD to reach the same training loss. The idea of MATCHA can be applied to any decentralized algorithm that involves a communication step with neighbors in a graph.

关键词： Convergence Training Topology network topology Delays Servers Optimization Decentralized SGD distributed Training Communication-efficient Methods

来源：评论

学校读者我要写书评

暂无评论

Neuromorphic Computing in Autonomous distributed Grid Control

Neuromorphic Computing in Autonomous Distributed Grid Contro...

引用

2024 IEEE International Conferences of AI, Science, Engineering, and Technology, AIxSET 2024

作者： Obert, James Trevizan, Rodrigo D. Chavez, Adrian Sandia National Labs AlbuquerqueNM United States

ISBN: (纸本)9798350390995

Due to the distributed nature of the electrical grid, intelligent, timely control of critical components such as volt-age regulators, capacitor banks and switches in a highly dynamic environment is extremely challenging. Modern day localized controllers in a distribution system such as remote ter-minal units (RTUs) and programmable logic units (PLCs) are often housed in a substation with multiple connections to several critical controls and are thus good candidates for hosting advanced localized control via deep reinforcement learning (DRL) or evolutionary learning. Additionally, it is feasible to greatly reduce the computational throughput and provide fine-grained control measurement and control using a spiking neural network (SNN) neural processing unit (NPU) architecture. In this paper we explore the use of a hybrid grid control algorithm composed of an optimized genetic algorithm (GA) and SNN controller which acquires optimal distribution system voltage control policies while a system is experiencing a cyber adver-sary's injected control noise. We discuss the computational advantages of the GA-SNN hybrid algorithm in the fine-grained autonomous control of a grid's bus voltages using a commercially available neuromorphic system on chip system (NSoC) with a mean rated power consumption of 1 mW and observed mean prediction time of 4 msecs per inference. © 2024 IEEE.

关键词： Deep reinforcement learning

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：