Article
Computer Science, Information Systems
Haozhao Wang, Zhihao Qu, Qihua Zhou, Haobo Zhang, Boyuan Luo, Wenchao Xu, Song Guo, Ruixuan Li
Summary: This article provides a comprehensive review of recent advances in accelerating the training of large machine learning models in IoT. It emphasizes optimization algorithms and distributed learning architectures, as well as computation hardware acceleration and communication optimization for collaborative training.
IEEE INTERNET OF THINGS JOURNAL
(2022)
Editorial Material
Multidisciplinary Sciences
Jeffrey M. Perkel
Summary: The new breed of notebooks combines data visualization and collaborative functionality, with the simplicity of a spreadsheet.
Article
Computer Science, Artificial Intelligence
Hugo Touvron, Piotr Bojanowski, Mathilde Caron, Matthieu Cord, Alaaeldin El-Nouby, Edouard Grave, Gautier Izacard, Armand Joulin, Gabriel Synnaeve, Jakob Verbeek, Herve Jegou
Summary: ResMLP is an image classification architecture that relies solely on multi-layer perceptrons. It achieves accurate and efficient results through a combination of linear layers and a two-layer feed-forward network, trained with modern strategies and data augmentation.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Editorial Material
Multidisciplinary Sciences
Aleksandra Piktus
Summary: Large language models are incorporating external tools like Wikipedia to enhance their accuracy in reasoning, which could lead to better fact-finding outcomes and online shopping experiences.
Article
Computer Science, Information Systems
Peng Sun, Yonggang Wen, Ruobing Han, Wansen Feng, Shengen Yan
Summary: Scaling out deep neural network (DNN) training is crucial for reducing model training time, but high communication overhead in distributed DNN training is a major performance bottleneck. In this study, we propose GradientFlow, a communication backend, and employ various network optimization techniques to tackle this problem. By integrating methods such as ring-based allreduce, mixed-precision training, and computation/communication overlap, as well as introducing lazy allreduce and coarse-grained sparse communication, we were able to achieve impressive speedup ratios when training AlexNet and ResNet-50 on the ImageNet dataset using multiple GPUs.
IEEE TRANSACTIONS ON BIG DATA
(2022)
Review
Chemistry, Physical
Hongyi Xu, Juner Zhu, Donal P. Finegan, Hongbo Zhao, Xuekun Lu, Wei Li, Nathaniel Hoffman, Antonio Bertei, Paul Shearing, Martin Z. Bazant
Summary: The electrochemical and mechanical properties of lithium-ion battery materials heavily rely on their 3D microstructure characteristics. A quantitative understanding of the role played by stochastic microstructures is crucial for predicting material properties and guiding synthesis processes. Tailoring microstructure morphology is also a viable way to achieve optimal electrochemical and mechanical performance of lithium-ion cells. This review presents spatially and temporally resolved imaging of microstructure and electrochemical phenomena, microstructure statistical characterization and stochastic reconstruction, microstructure-resolved modeling for property prediction, and machine learning for microstructure design to facilitate the establishment of microstructure-resolved modeling and design methods. Perspectives on the unresolved challenges and opportunities in applying experimental data, modeling, and machine learning to improve understanding of materials and identify paths toward enhanced performance of lithium-ion cells are also discussed.
ADVANCED ENERGY MATERIALS
(2021)
Article
Computer Science, Theory & Methods
Qingyang Duan, Chao Peng, Zeqin Wang, Yuedong Xu, Shaoteng Liu, Jun Wu, John C. S. Lui
Summary: This article proposes a simple transport layer scheduler called Mercury, which moves the priority scheduling to the transport layer at the packet granularity to achieve optimal overlap between communication and computation. Experimental results show that Mercury can accelerate training by up to 130% compared to the classical PS architecture, and 104% compared to state-of-the-art tensor partitioning methods.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Xupeng Miao, Lingxiao Ma, Zhi Yang, Yingxia Shao, Bin Cui, Lele Yu, Jiawei Jiang
Summary: This paper proposes a high-efficient GPU training framework called cuWide for large-scale wide models. By adopting a new flow-based training schema, cuWide utilizes the memory hierarchy of GPU to reduce communication with GPU global memory. Optimization strategies such as 2D partition of mini-batch and spatial-temporal caching mechanisms further improve the performance.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2022)
Article
Computer Science, Artificial Intelligence
Pan Mu, Zhu Liu, Yaohua Liu, Risheng Liu, Xin Fan
Summary: In this paper, a model-guided triple-level optimization framework, named TMICS, is proposed for video deraining. The framework involves designing collaborative structures, searching inter-frame information, and automatically discovering rain streaks removal architectures, leading to significant improvements in fidelity and temporal consistency.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Yanan Sun, Xian Sun, Yuhan Fang, Gary G. Yen, Yuqiao Liu
Summary: Evolutionary neural architecture search (ENAS) and performance predictors are discussed in their application in designing deep neural networks. A new training protocol is proposed to optimize performance prediction, with experimental results demonstrating significant improvement in prediction accuracy compared to traditional protocols.
IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION
(2021)
Article
Computer Science, Software Engineering
Mickael Sereno, Xiyao Wang, Lonni Besancon, Michael J. Mcguffin, Tobias Isenberg
Summary: This paper presents the state of existing work at the intersection of augmented reality and computer-supported collaborative work (AR-CSCW), categorizing 65 papers and deriving design considerations for collaborative AR environments. It also identifies under-explored research topics, providing useful information for newcomers, interested readers, and domain experts.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS
(2022)
Article
Computer Science, Theory & Methods
Nan Wu, Yuan Xie
Summary: This article explores the application of machine learning in computer architecture and system design, presenting a comprehensive review of the work in this field. It discusses the role of machine learning techniques in architecture/system design and summarizes the problems that can be solved using these techniques. The article also provides insights into the potential directions and opportunities for applying machine learning in this domain.
ACM COMPUTING SURVEYS
(2023)
Article
Engineering, Electrical & Electronic
Yin Tang, Qi Teng, Lei Zhang, Fuhong Min, Jun He
Summary: The research introduces a lightweight CNN model using Lego filters for HAR tasks, achieving higher accuracy and reducing memory and computation cost. Experimental results show that this Lego CNN model is smaller, faster, and more accurate than traditional CNN. Moreover, the model does not rely on special network structures and has application potential in ubiquitous and wearable computing.
IEEE SENSORS JOURNAL
(2021)
Article
Computer Science, Hardware & Architecture
Jianhang Tang, Jiangtian Nie, Yang Zhang, Yiqun Duan, Zehui Xiong, Dusit Niyato
Summary: This paper proposes an air-ground collaborative edge intelligence framework to enable persistent and ubiquitous AI services, addressing the challenge of resource sharing between terrestrial MEC networks and UAVs. The paper discusses three air-ground collaboration schemes and a novel machine learning model caching approach, and provides simulation results to demonstrate the effectiveness of the proposed algorithm.
Article
Engineering, Mechanical
Tianju Chen, Mark C. Messner
Summary: Accurate constitutive models are necessary for high temperature design to describe material inelastic deformation and failure behavior. The pyopmat package presented in this study is an open source framework that utilizes machine learning techniques to calibrate constitutive models against experiment data under various loading conditions. The package calculates the exact gradient of the model response with respect to the parameters using automatic differentiation and the adjoint method. The efficiency and accuracy of the package are demonstrated through example problems with synthetic data and actual high temperature creep-fatigue test data.
INTERNATIONAL JOURNAL OF PLASTICITY
(2023)
Editorial Material
Computer Science, Artificial Intelligence
Xiaochun Cheng, Chengqi Zhang, Yi Qian, Moayad Aloqaily, Yang Xiao
INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS
(2021)
Article
Computer Science, Theory & Methods
Deniz Ozsoyeller, Oznur Ozkasap, Moayad Aloqaily
Summary: This study focuses on the problem of asynchronous rendezvous search with multiple mobile agents in the plane. The algorithm m-RENDEZVOUS is proposed to achieve deterministic rendezvous by utilizing waiting robots to break symmetry. Extensive simulations were conducted to verify the performance of the algorithm, in addition to theoretical results.
FUTURE GENERATION COMPUTER SYSTEMS-THE INTERNATIONAL JOURNAL OF ESCIENCE
(2022)
Article
Engineering, Civil
Moayad Aloqaily, Haya Elayan, Mohsen Guizani
Summary: The advancement of wireless connectivity in smart cities enhances connections between key elements, and the federated intelligent health monitoring systems in autonomous vehicles contribute to improving quality of life. This study proposes C-HealthIER, a cooperative health intelligent emergency response system that monitors passengers' health and conducts cooperative behavior to reduce emergency treatment time and distance by sharing information between vehicles and infrastructure.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2023)
Article
Engineering, Civil
Moayad Aloqaily, Ismaeel Al Ridhawi, Mohsen Guizani
Summary: This article introduces a solution that integrates the capabilities of UAVs and UGVs to provide intelligent connectivity and services to aerial and ground connected devices. It ensures continuous power availability to UAVs through cooperative processing, and employs Federated Learning and Blockchain technology for accurate and secure service provisioning. Extensive simulations show significant improvements in connectivity, service availability, and UAV energy enhancements compared to traditional communication techniques.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Computer Science, Information Systems
Moayad Aloqaily, Ouns Bouachir, Oznur Ozkasap, Faizan Safdar Ali
Summary: The development of intelligent cities involves increasing local energy generation through renewable resources, allowing for energy trading between local producers and consumers. A hybrid approach involving both local and external prosumers is necessary, facilitated by a centralized energy trading system like SynergyGrids, which reduces energy costs and grid loads. This hybrid energy trading platform utilizing smart contracts shows promising results in lowering costs and improving energy management.
PEER-TO-PEER NETWORKING AND APPLICATIONS
(2022)
Article
Computer Science, Hardware & Architecture
Moayad Aloqaily, Rasheed Hussain, Deena Khalaf, Dana Slehat, Alma Oracevic
Summary: Unmanned aerial vehicles (UAVs) and autonomous vehicle (AV) technologies complement each other by providing support and addressing each other's limitations at the service and application level. This article focuses on the role of futuristic technologies, such as blockchain and artificial intelligence, in securing critical environments and discusses the challenges and opportunities for research and development.
IEEE CONSUMER ELECTRONICS MAGAZINE
(2022)
Article
Computer Science, Theory & Methods
Tan Guo, Keping Yu, Moayad Aloqaily, Shaohua Wan
Summary: The Artificial Intelligence Internet of Things (AIoT) is an emerging concept that aims to connect intelligent things for efficient intercommunication. This paper presents a prior-dependent graph (PDG) construction method to model and discover complex relations in data, enabling high-efficiency data clustering and dimensionality reduction. Experimental results demonstrate that the PDG model achieves substantial performance compared to existing graph learning models.
FUTURE GENERATION COMPUTER SYSTEMS-THE INTERNATIONAL JOURNAL OF ESCIENCE
(2022)
Editorial Material
Computer Science, Hardware & Architecture
Moayad Aloqaily, Kobbane Abdellatif, Feng Yan
MOBILE NETWORKS & APPLICATIONS
(2022)
Article
Construction & Building Technology
Moayad Aloqaily, Ouns Bouachir, Ismaeel Al Ridhawi, Anthony Tzes
Summary: This article presents a UAV-supported vehicular network solution that considers power and coverage limitations of UAVs to support sustainable smart cities. The solution dynamically adjusts the UAV height based on predictive optimization to achieve optimal communication coverage. The proposed model is evaluated through implementation and simulation, showing accurate prediction of vehicle traffic patterns and adherence to QoS requirements for network coverage.
SUSTAINABLE CITIES AND SOCIETY
(2022)
Article
Automation & Control Systems
Anran Du, Yicheng Shen, Qinzi Zhang, Lewis Tseng, Moayad Aloqaily
Summary: Industry 5.0 is emerging as a result of advancements in networking and communication technologies, artificial intelligence, distributed computing, and beyond 5G. This article proposes a framework to integrate federated learning, industrial edge computing, and Byzantine-tolerant machine learning, and introduces a novel Byzantine-tolerant federated learning algorithm called CRACAU.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2022)
Review
Computer Science, Information Systems
Vahideh Hayyolalam, Moayad Aloqaily, Oznur Ozkasap, Mohsen Guizani
Summary: With the rapid growth of edge-assisted solutions in IoT networks, connected healthcare increasingly relies on such solutions. These systems employ novel technologies, such as IoT, edge computing, and AI, to transform conventional health systems into more effective, personalized intelligent systems. However, these systems face restrictions and require new policies. Edge computing, achieved by moving computation and processing closer to data sources and end-users, can reduce latency, bandwidth usage, and energy consumption.
IEEE INTERNET OF THINGS JOURNAL
(2022)
Article
Telecommunications
Ouns Bouachir, Moayad Aloqaily, Oznur Ozkasap, Faizan Ali
Summary: This paper introduces a new P2P energy trading and sharing platform called FederatedGrids, which utilizes blockchain and federated learning technology to enable autonomous activities, trust, and privacy protection. The platform enables energy sharing, predicts future energy demand through smart contracts and federated learning, and automatically switches between trading and sharing.
IEEE TRANSACTIONS ON GREEN COMMUNICATIONS AND NETWORKING
(2022)
Article
Computer Science, Theory & Methods
Ismaeel Al Ridhawi, Ouns Bouachir, Moayad Aloqaily, Azzedine Boukerche
Summary: IoT systems have greatly advanced with the support of ML and AI solutions, playing a crucial role in smart city services. A cooperative intelligent and secure IoT framework is necessary to achieve true smart city experiences, considering the heterogeneous wireless networks and abundance of IoT devices.
ACM COMPUTING SURVEYS
(2022)
Proceedings Paper
Computer Science, Hardware & Architecture
Haya Elayan, Moayad Aloqaily, Haythem Bany Salameh, Mohsen Guizani
Summary: Recent technological advances have reshaped modern transportation systems, with AI and B5G networks increasing automation levels. A proposed cooperative healthcare emergency response framework aims to minimize emergency treatment time for passengers with abnormal health conditions, utilizing in-vehicle intelligent health monitoring and local networks. Simulation results show significant reductions in emergency treatment time, hospital waiting time, and travel time compared to autopilot approaches.
PROCEEDINGS OF THE IEEE 46TH CONFERENCE ON LOCAL COMPUTER NETWORKS (LCN 2021)
(2021)