Related references
Note: Only part of the references are listed.
Article
Computer Science, Hardware & Architecture
Xizi Chen et al.
Summary: In this work, a model compression method based on a novel weight permutation scheme is proposed to fully exploit the fine-grained weight sparsity in hardware design. The optimal arrangement of the weight matrix is obtained through permutation, and the sparse weight matrix is further compressed to a small and dense format to make full use of hardware resources. In addition, a more fine-grained subword-level pruning is proposed to further improve the compression performance.
IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS
(2023)
Article
Computer Science, Information Systems
Chunhua Zhao et al.
Summary: This article proposes a multilayer joint optimization of packet size and adaptive transmission scheduling (ATS) to optimize the energy consumption and prolong the network lifetime of wireless sensor networks (WSNs) for mechanical vibration monitoring (MVM). The packet size is optimized based on the lowest energy consumption (PSO-LEC) model, and ATS based on the time division multiple access (TDMA/ATS) method is used to optimize data collision and transmission delay problems. Experimental results show that the proposed scheme significantly improves transmission reliability, energy efficiency, and network life for MVM.
IEEE INTERNET OF THINGS JOURNAL
(2023)
Article
Automation & Control Systems
Chunhua Zhao et al.
Summary: This article proposes a novel multilevel adaptive near-lossless compression method in edge collaborative WSN for mechanical vibration monitoring, addressing the challenges of limited storage and computational resources as well as high delay in transmitting massive vibration data. The proposed method characterizes mechanical fault feature information accurately in low storage space and integrates edge computing technique to reduce the pressure on data center servers. Comprehensive experiments demonstrate that the proposed approach achieves high-precision data reconstruction and feature detection, significantly improving the computational power and transmission efficiency of WSN.
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
(2023)
Article
Computer Science, Information Systems
Jose Jaena Mari Ople et al.
Summary: Generative Adversarial Networks (GAN) is a powerful machine learning method for image generation. However, its large memory footprint limits its practical applications for resource-constrained devices. To address this, we propose AdjustableGAN, a method that can compress pretrained GAN models to different compression ratios using genetic algorithms. Experimental results show that our method can reliably compress generative networks with minimal accuracy drop compared to other state-of-the-art compression algorithms.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Computer Science, Artificial Intelligence
Karan Bajaj et al.
Summary: IoT applications and services are becoming pervasive in various domains, requiring high-speed computation and processing of vast amounts of data; computation offloading frameworks play a crucial role in meeting performance requirements, with context-based offloading being essential to achieve optimum results.
COMPLEX & INTELLIGENT SYSTEMS
(2022)
Article
Engineering, Electrical & Electronic
Xiuwen Fu et al.
Summary: This study proposes a directional load metric to investigate the cascade failure process in the Internet of Things (IoT). Experimental results show that the network in the local routing mode is more reliable, and increasing the number of gateways significantly improves the reliability of the IoT. These findings provide guidance for users in building a reliable IoT system against cascade failures.
IEEE SENSORS JOURNAL
(2022)
Article
Computer Science, Information Systems
Pavana Prakash et al.
Summary: In this article, we propose GWEP, a model compression-based federated learning method, which leverages joint quantization and model pruning to enable efficient deep learning model training on resource-constrained IoT devices. Through empirical evaluations, we demonstrate that our approach significantly outperforms baseline algorithms in terms of speed, communication rounds, compression rate, and other performance metrics.
IEEE INTERNET OF THINGS JOURNAL
(2022)
Article
Automation & Control Systems
Honghai Huang et al.
Summary: This article proposes a new method for gearbox fault diagnosis using multisensor fusion. The residual gated dynamic sparse network is used to improve feature learning and fusion ability. Experimental results and engineering application show that this method is more effective than others under noise interference.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2022)
Article
Computer Science, Artificial Intelligence
Ye Liu et al.
Summary: This paper proposes a new algorithm to compress deep neural networks by considering both nonlinear response and multilinear low-rank constraint in the kernel tensor. The effectiveness of the proposed method is demonstrated on various CNNs and datasets, achieving a significantly reduced model size without significant loss in accuracy. Compared to existing methods, the proposed method outperforms in compression performance.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Information Systems
Enrico Russo et al.
Summary: This article proposes a compression technique called LineCompress, which aims to reduce memory and communication overhead on resource-constrained IoT devices, thus improving the inference speed and energy efficiency of deep neural networks.
IEEE INTERNET OF THINGS JOURNAL
(2022)
Article
Engineering, Electrical & Electronic
Yi Huang et al.
Summary: This paper proposes a beacon synchronization-based multi-channel with dynamic time slot assignment method for wireless sensor networks (WSNs) in mechanical vibration monitoring. The method separates the network into two opposing groups based on working state and allocates channels based on interference minimization. It also introduces a dynamic timeslot assignment method to reduce channel collision. Experimental results show that the proposed methods effectively schedule the network for data transmission via multi-channel, outperforming CSMA/CA and TDMA.
IEEE SENSORS JOURNAL
(2022)
Article
Computer Science, Information Systems
Lanlan Rui et al.
Summary: This paper proposes a model compression algorithm based on model pruning and model clustering in smart network maintenance, which can effectively compress the width and depth of the model while maintaining the accuracy of the model, saving storage space, and simplifying calculation.
IEEE TRANSACTIONS ON NETWORK AND SERVICE MANAGEMENT
(2022)
Article
Engineering, Industrial
Xiuwen Fu et al.
Summary: This paper develops a cascading model that takes into account the characteristics of edge-assisted IoTs and investigates the impact of cascading failure on their reliability. The experimental results show that isolation is the main cause of performance degradation, and link density and cluster head ratio have a positive correlation with cascading reliability.
RELIABILITY ENGINEERING & SYSTEM SAFETY
(2022)
Article
Chemistry, Analytical
Ming Zhao et al.
Summary: In this paper, a strategy for refining neural network filters is proposed to balance the tradeoff between the deep-learning module-compression ratio and the module accuracy. Experimental results show that the method can significantly reduce the computational effort of the neural network and compress the number of parameters in the model.
Article
Telecommunications
Rakesh Ranjan Swain et al.
Summary: This article covers the basics of fault diagnosis in wireless sensor networks, including fault diagnosis terminology, sensor fault classification, causes and effects, and fault diagnosis performance metrics. It reviews the state-of-the-art fault diagnosis approaches for wireless sensor networks and discusses the benefits and drawbacks of existing automated fault detection and diagnosis approaches.
WIRELESS PERSONAL COMMUNICATIONS
(2022)
Article
Computer Science, Information Systems
Cheng Dai et al.
Summary: Deep learning has been shown to be an effective method for feature encoding in intelligent applications, but its large memory consumption hinders its deployment on embedded systems with limited resources. To address this challenge, we propose a novel compression technique that reduces memory size and inference complexity while maintaining a small drop in accuracy. We introduce improved pruning and decomposition methods, and extensive experiments demonstrate the superior performance of our approach compared to existing methods in terms of accuracy, parameter compression, and reduction in floating-point operations.
IEEE INTERNET OF THINGS JOURNAL
(2022)
Article
Automation & Control Systems
Jiedi Sun et al.
Summary: This paper proposes a multi-hierarchical compression method to reduce the complexity of deep neural networks in industrial environments with constrained resources, while achieving intelligent fault diagnosis without a decrease in recognition accuracy.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2022)
Article
Engineering, Electrical & Electronic
Cristian Wisultschew et al.
Summary: The shift of computing from the cloud to the edge of the Internet of Things is influencing deep learning applications. Moving intelligence closer to the point of need has advantages in terms of performance, power consumption, security, and privacy. However, dealing with massive data generated by data sources such as LIDAR sensors becomes challenging for edge devices, especially when heavy processing algorithms like deep neural networks are selected. This work demonstrates the feasibility of processing point cloud-based sensors at the edge using deep neural networks, thanks to new devices with high computing capacity and reduced power consumption. Evaluation of different edge processing architectures shows that neural accelerators with integrated host CPUs provide the best trade-off between power consumption and performance, making them an ideal solution for IoT applications at the edge level.
IEEE SENSORS JOURNAL
(2022)
Article
Computer Science, Theory & Methods
Shun Liu et al.
Summary: This paper proposes a solution to the task offloading problem on Vehicular Edge Computing (VEC) servers, which utilizes the Asynchronous Advantage Actor-Critic (A3C) algorithm to minimize task completion time and reduce the delay of big data processing.
JOURNAL OF PARALLEL AND DISTRIBUTED COMPUTING
(2022)
Article
Telecommunications
Yogesh Tripathi et al.
Summary: The duty cycling process is a promising approach for designing routing protocols in resource-constrained Wireless Sensor Networks (WSNs). It improves network lifetime and transmission efficiency compared to conventional protocols. This study proposes a routing protocol that divides the duty cycle into slots to minimize idle listening and uses the opportunistic concept to minimize sender waiting. The simulation analysis shows that the proposed scheme outperforms existing ones in terms of transmission delay, energy consumption, and network throughput.
DIGITAL COMMUNICATIONS AND NETWORKS
(2022)
Article
Computer Science, Information Systems
Anthony Berthelier et al.
Summary: This paper surveys methods suitable for porting deep neural networks on resource-limited devices, especially for smart cameras, which can be roughly divided into compression techniques and architecture optimization. Compression techniques include knowledge distillation, pruning, quantization, hashing, reduction of numerical precision and binarization, while architecture optimization focuses on enhancing network structures and neural architecture search techniques.
JOURNAL OF SIGNAL PROCESSING SYSTEMS FOR SIGNAL IMAGE AND VIDEO TECHNOLOGY
(2021)
Article
Engineering, Electrical & Electronic
Jinyang Guo et al.
Summary: The proposed Progressive Channel Pruning (PCP) framework accelerates Convolutional Neural Networks (CNNs) by iteratively pruning a small number of channels from selected layers using a three-step pipeline. A greedy strategy is used to automatically select layers to minimize overall accuracy drop after pruning, leading to superior performance in supervised and transfer learning settings.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2021)
Article
Computer Science, Hardware & Architecture
Cheng Gong et al.
Summary: This paper introduces a novel quantization solution called VecQ, which decomposes the minimization of Direct Quantization Loss (DQL) into two independent optimization processes using a new metric Vector Loss. By evaluating on various datasets and models, the results demonstrate that the proposed solution is more accurate and effective than existing methods, with more flexible bitwidth support.
IEEE TRANSACTIONS ON COMPUTERS
(2021)
Article
Computer Science, Information Systems
Jianping Zhu et al.
Summary: Mobile-RadarNet is an efficient convolutional neural network architecture that achieves high classification accuracy while keeping computational complexity extremely low, making it suitable for deployment in mobile devices.
IEEE INTERNET OF THINGS JOURNAL
(2021)
Article
Computer Science, Information Systems
JinYi Yoon et al.
Summary: EdgePipe is a deep learning framework that utilizes a combination of model parallelism and pipeline training, improving resource utilization on edge devices while addressing volatility issues in wireless links and device functionality.
IEEE INTERNET OF THINGS JOURNAL
(2021)
Article
Acoustics
Ke Tan et al.
Summary: The study proposes two compression pipelines to reduce the size of DNN-based speech enhancement models using sparse regularization, iterative pruning, and clustering-based quantization. Experimental results show that this approach reduces model sizes while maintaining enhancement performance, especially excelling in speaker separation tasks.
IEEE-ACM TRANSACTIONS ON AUDIO SPEECH AND LANGUAGE PROCESSING
(2021)
Article
Computer Science, Information Systems
Xiuwen Fu et al.
Summary: The proposed Trend Correlation based Fault Detection strategy (TCFD) can detect faulty sensor nodes by analyzing trend correlation and neighboring nodes' median value. A fault detection self-starting mechanism is designed to avoid excessive routing overhead caused by frequent fault detection. TCFD's detection results are based on historical data, reducing the impact of fault detection time on accuracy and achieving better fault detection accuracy for common fault types.
Article
Engineering, Electrical & Electronic
Ching-Hu Lu et al.
IEEE SENSORS JOURNAL
(2020)
Article
Computer Science, Artificial Intelligence
Y Rubner et al.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2000)
