Pub Date : 2024-02-05DOI: 10.1109/TCC.2024.3361858
Yazhou Yuan;Shicong Gao;Ziteng Zhang;Wenye Wang;Zhezhuang Xu;Zhixin Liu
With the rapid development of artificial intelligence and Unmanned Aerial Vehicle (UAV) technology, AI-based UAVs are increasingly utilized in various industrial and civilian applications. This paper presents a distributed Edge-Cloud collaborative framework for UAV object detection, aiming to achieve real-time and accurate detection of ground moving targets. The framework incorporates an Edge-Embedded Lightweight ( ${{text{E}}^{2}}text{L}$ ) object algorithm with an attention mechanism, enabling real-time object detection on edge-side embedded devices while maintaining high accuracy. Additionally, a decision-making mechanism based on fuzzy neural network facilitates adaptive task allocation between the edge-side and cloud-side. Experimental results demonstrate the improved running rate of the proposed algorithm compared to YOLOv4 on the edge-side NVIDIA Jetson Xavier NX, and the superior performance of the distributed Edge-Cloud collaborative framework over traditional edge computing or cloud computing algorithms in terms of speed and accuracy
{"title":"Edge-Cloud Collaborative UAV Object Detection: Edge-Embedded Lightweight Algorithm Design and Task Offloading Using Fuzzy Neural Network","authors":"Yazhou Yuan;Shicong Gao;Ziteng Zhang;Wenye Wang;Zhezhuang Xu;Zhixin Liu","doi":"10.1109/TCC.2024.3361858","DOIUrl":"10.1109/TCC.2024.3361858","url":null,"abstract":"With the rapid development of artificial intelligence and Unmanned Aerial Vehicle (UAV) technology, AI-based UAVs are increasingly utilized in various industrial and civilian applications. This paper presents a distributed Edge-Cloud collaborative framework for UAV object detection, aiming to achieve real-time and accurate detection of ground moving targets. The framework incorporates an Edge-Embedded Lightweight (\u0000<inline-formula><tex-math>${{text{E}}^{2}}text{L}$</tex-math></inline-formula>\u0000) object algorithm with an attention mechanism, enabling real-time object detection on edge-side embedded devices while maintaining high accuracy. Additionally, a decision-making mechanism based on fuzzy neural network facilitates adaptive task allocation between the edge-side and cloud-side. Experimental results demonstrate the improved running rate of the proposed algorithm compared to YOLOv4 on the edge-side NVIDIA Jetson Xavier NX, and the superior performance of the distributed Edge-Cloud collaborative framework over traditional edge computing or cloud computing algorithms in terms of speed and accuracy","PeriodicalId":13202,"journal":{"name":"IEEE Transactions on Cloud Computing","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2024-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139951290","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-02-01DOI: 10.1109/TCC.2024.3361070
Bruno Guindani;Danilo Ardagna;Alessandra Guglielmi;Roberto Rocco;Gianluca Palermo
Bayesian Optimization (BO) is an efficient method for finding optimal cloud configurations for several types of applications. On the other hand, Machine Learning (ML) can provide helpful knowledge about the application at hand thanks to its predicting capabilities. This work proposes a general approach based on BO, which integrates elements from ML techniques in multiple ways, to find an optimal configuration of recurring jobs running in public and private cloud environments, possibly subject to black-box constraints, e.g., application execution time or accuracy. We test our approach by considering several use cases, including edge computing, scientific computing, and Big Data applications. Results show that our solution outperforms other state-of-the-art black-box techniques, including classical autotuning and BO- and ML-based algorithms, reducing the number of unfeasible executions and corresponding costs up to 2–4 times.
贝叶斯优化（BO）是一种高效的方法，可为多种类型的应用找到最佳云配置。另一方面，机器学习（ML）凭借其预测能力，可以提供有关当前应用的有用知识。这项工作提出了一种基于 BO 的通用方法，该方法以多种方式集成了 ML 技术的各种元素，可为在公共和私有云环境中运行的重复性工作找到最佳配置，可能会受到黑盒子约束，例如应用程序的执行时间或准确性。我们考虑了多个使用案例，包括边缘计算、科学计算和大数据应用，对我们的方法进行了测试。结果表明，我们的解决方案优于其他最先进的黑盒技术，包括经典的自动调整以及基于 BO 和 ML 的算法，可将不可行的执行次数和相应的成本最多减少 2-4 倍。
{"title":"Integrating Bayesian Optimization and Machine Learning for the Optimal Configuration of Cloud Systems","authors":"Bruno Guindani;Danilo Ardagna;Alessandra Guglielmi;Roberto Rocco;Gianluca Palermo","doi":"10.1109/TCC.2024.3361070","DOIUrl":"10.1109/TCC.2024.3361070","url":null,"abstract":"Bayesian Optimization (BO) is an efficient method for finding optimal cloud configurations for several types of applications. On the other hand, Machine Learning (ML) can provide helpful knowledge about the application at hand thanks to its predicting capabilities. This work proposes a general approach based on BO, which integrates elements from ML techniques in multiple ways, to find an optimal configuration of recurring jobs running in public and private cloud environments, possibly subject to black-box constraints, e.g., application execution time or accuracy. We test our approach by considering several use cases, including edge computing, scientific computing, and Big Data applications. Results show that our solution outperforms other state-of-the-art black-box techniques, including classical autotuning and BO- and ML-based algorithms, reducing the number of unfeasible executions and corresponding costs up to 2–4 times.","PeriodicalId":13202,"journal":{"name":"IEEE Transactions on Cloud Computing","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2024-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10418550","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139951560","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-31DOI: 10.1109/TCC.2024.3360443
Che Chen;Shimin Gong;Wenjie Zhang;Yifeng Zheng;Yeo Chai Kiat
Mobile edge computing (MEC) is viewed as a promising technology to address the challenges of intensive computing demands in hotspots (HSs). In this article, we consider a unmanned aerial vehicle (UAV)-assisted backscattering MEC system. The UAVs can fly from parking aprons to HSs, providing energy to HSs via RF beamforming and collecting data from wireless users in HSs through backscattering. We aim to maximize the long-term utility of all HSs, subject to the stability of the HSs’ energy queues. This problem is a joint optimization of the data offloading decision and contract design that should be adaptive to the users’ random task demands and the time-varying wireless channel conditions. A deep reinforcement learning based contract incentive (DRLCI) strategy is proposed to solve this problem in two steps. First, we use deep Q-network (DQN) algorithm to update the HSs’ offloading decisions according to the changing network environment. Second, to motivate the UAVs to participate in resource sharing, a contract specific to each type of UAVs has been designed, utilizing Lagrangian multiplier method to approach the optimal contract. Simulation results show the feasibility and efficiency of the proposed strategy, demonstrating a better performance than the natural DQN and Double-DQN algorithms.
{"title":"DRL-Based Contract Incentive for Wireless-Powered and UAV-Assisted Backscattering MEC System","authors":"Che Chen;Shimin Gong;Wenjie Zhang;Yifeng Zheng;Yeo Chai Kiat","doi":"10.1109/TCC.2024.3360443","DOIUrl":"10.1109/TCC.2024.3360443","url":null,"abstract":"Mobile edge computing (MEC) is viewed as a promising technology to address the challenges of intensive computing demands in hotspots (HSs). In this article, we consider a unmanned aerial vehicle (UAV)-assisted backscattering MEC system. The UAVs can fly from parking aprons to HSs, providing energy to HSs via RF beamforming and collecting data from wireless users in HSs through backscattering. We aim to maximize the long-term utility of all HSs, subject to the stability of the HSs’ energy queues. This problem is a joint optimization of the data offloading decision and contract design that should be adaptive to the users’ random task demands and the time-varying wireless channel conditions. A deep reinforcement learning based contract incentive (DRLCI) strategy is proposed to solve this problem in two steps. First, we use deep Q-network (DQN) algorithm to update the HSs’ offloading decisions according to the changing network environment. Second, to motivate the UAVs to participate in resource sharing, a contract specific to each type of UAVs has been designed, utilizing Lagrangian multiplier method to approach the optimal contract. Simulation results show the feasibility and efficiency of the proposed strategy, demonstrating a better performance than the natural DQN and Double-DQN algorithms.","PeriodicalId":13202,"journal":{"name":"IEEE Transactions on Cloud Computing","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2024-01-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139951383","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-30DOI: 10.1109/TCC.2024.3360098
Yafeng Chen;Axin Wu;Yuer Yang;Xiangjun Xin;Chang Song
Private set intersection cardinality (PSI-CA) allows two parties to learn the size of the intersection between two private sets without revealing other additional information, which is a promising technique to solve privacy concerns in contact tracing. Efficient PSI protocols typically use oblivious transfer, involving multiple rounds of interaction and leading to heavy local computation overheads and protocol delays, especially when interacting with many receivers. Cloud-assisted PSI-CA is a better solution as it relieves participants’ burdens of computation and communication. However, cloud servers may return incorrect or incomplete results for some reason, leading to an incorrectness issue. At present, to our knowledge, existing cloud-assisted PSI-CA protocols cannot address such a concern. To address this, we propose two specific verifiable cloud-assisted PSI-CA protocols: one based on a two-server protocol and the other on a single-server protocol. Further, we employ Cuckoo hashing to optimize these two protocols, enabling the receiver's computational costs independent of the size of the sender's set. We also prove the security of the protocols and implement them. Finally, we analyze and discuss their performance demonstrating that the single-server verifiable PSI-CA protocol does not introduce significant computation or communication costs while adding functionalities.
{"title":"Efficient Verifiable Cloud-Assisted PSI Cardinality for Privacy-Preserving Contact Tracing","authors":"Yafeng Chen;Axin Wu;Yuer Yang;Xiangjun Xin;Chang Song","doi":"10.1109/TCC.2024.3360098","DOIUrl":"10.1109/TCC.2024.3360098","url":null,"abstract":"Private set intersection cardinality (PSI-CA) allows two parties to learn the size of the intersection between two private sets without revealing other additional information, which is a promising technique to solve privacy concerns in contact tracing. Efficient PSI protocols typically use oblivious transfer, involving multiple rounds of interaction and leading to heavy local computation overheads and protocol delays, especially when interacting with many receivers. Cloud-assisted PSI-CA is a better solution as it relieves participants’ burdens of computation and communication. However, cloud servers may return incorrect or incomplete results for some reason, leading to an incorrectness issue. At present, to our knowledge, existing cloud-assisted PSI-CA protocols cannot address such a concern. To address this, we propose two specific verifiable cloud-assisted PSI-CA protocols: one based on a two-server protocol and the other on a single-server protocol. Further, we employ Cuckoo hashing to optimize these two protocols, enabling the receiver's computational costs independent of the size of the sender's set. We also prove the security of the protocols and implement them. Finally, we analyze and discuss their performance demonstrating that the single-server verifiable PSI-CA protocol does not introduce significant computation or communication costs while adding functionalities.","PeriodicalId":13202,"journal":{"name":"IEEE Transactions on Cloud Computing","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2024-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139951366","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-29DOI: 10.1109/TCC.2024.3358823
Bartosz Żurkowski;Krzysztof Zieliński
Root cause analysis (RCA) is a critical component in maintaining the reliability and performance of modern cloud applications. However, due to the inherent complexity of cloud environments, traditional RCA techniques become insufficient in supporting system administrators in daily incident response routines. This article presents an RCA solution specifically designed for cloud applications, capable of pinpointing failure root causes and recreating complete fault trajectories from the root cause to the effect. The novelty of our approach lies in approximating causal symptom dependencies by synergizing several symptom correlation methods that assess symptoms in terms of structural, semantic, and temporal aspects. The solution integrates statistical methods with system structure and behavior mining, offering a more comprehensive analysis than existing techniques. Based on these concepts, in this work, we provide definitions and construction algorithms for RCA model structures used in the inference, propose a symptom correlation framework encompassing essential elements of symptom data analysis, and provide a detailed description of the elaborated root cause identification process. Functional evaluation on a live microservice-based system demonstrates the effectiveness of our approach in identifying root causes of complex failures across multiple cloud layers.
{"title":"Root Cause Analysis for Cloud-Native Applications","authors":"Bartosz Żurkowski;Krzysztof Zieliński","doi":"10.1109/TCC.2024.3358823","DOIUrl":"10.1109/TCC.2024.3358823","url":null,"abstract":"Root cause analysis (RCA) is a critical component in maintaining the reliability and performance of modern cloud applications. However, due to the inherent complexity of cloud environments, traditional RCA techniques become insufficient in supporting system administrators in daily incident response routines. This article presents an RCA solution specifically designed for cloud applications, capable of pinpointing failure root causes and recreating complete fault trajectories from the root cause to the effect. The novelty of our approach lies in approximating causal symptom dependencies by synergizing several symptom correlation methods that assess symptoms in terms of structural, semantic, and temporal aspects. The solution integrates statistical methods with system structure and behavior mining, offering a more comprehensive analysis than existing techniques. Based on these concepts, in this work, we provide definitions and construction algorithms for RCA model structures used in the inference, propose a symptom correlation framework encompassing essential elements of symptom data analysis, and provide a detailed description of the elaborated root cause identification process. Functional evaluation on a live microservice-based system demonstrates the effectiveness of our approach in identifying root causes of complex failures across multiple cloud layers.","PeriodicalId":13202,"journal":{"name":"IEEE Transactions on Cloud Computing","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2024-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139951361","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-23DOI: 10.1109/TCC.2024.3357595
Ahmed M. Abdelmoniem;Brahim Bensaou
In data-centers, the composite origin and bursty nature of traffic, the small bandwidth-delay product and the tiny switch buffers lead to unusual congestion patterns that are not handled well by traditional end-to-end congestion control mechanisms such as those deployed in TCP. Existing works address the problem by modifying TCP to adapt it to the idiosyncrasies of data-centers. While this is feasible in private environments, it remains almost impossible to achieve practically in public multi-tenant clouds where a multitude of operating systems and thus congestion control protocols co-exist. In this work, we design a simple switch-based active queue management scheme to deal with such congestion issues adequately. Our approach requires no modification to TCP which enables seamless deployment in public data-centers via switch firmware updates. We present a simple analysis to show the stability and effectiveness of our approach, then discuss the real implementations in software and hardware on the NetFPGA platform. Numerical results from ns-2 simulation and experimental results from a small testbed cluster demonstrate the effectiveness of our approach in achieving high overall throughput, good fairness, smaller flow completion times (FCT) for short-lived flows, and a significant reduction in the tail of the FCT distribution.
{"title":"Alleviating Congestion via Switch Design for Fair Buffer Allocation in Datacenters","authors":"Ahmed M. Abdelmoniem;Brahim Bensaou","doi":"10.1109/TCC.2024.3357595","DOIUrl":"10.1109/TCC.2024.3357595","url":null,"abstract":"In data-centers, the composite origin and bursty nature of traffic, the small bandwidth-delay product and the tiny switch buffers lead to unusual congestion patterns that are not handled well by traditional end-to-end congestion control mechanisms such as those deployed in TCP. Existing works address the problem by modifying TCP to adapt it to the idiosyncrasies of data-centers. While this is feasible in private environments, it remains almost impossible to achieve practically in public multi-tenant clouds where a multitude of operating systems and thus congestion control protocols co-exist. In this work, we design a simple switch-based active queue management scheme to deal with such congestion issues adequately. Our approach requires no modification to TCP which enables seamless deployment in public data-centers via switch firmware updates. We present a simple analysis to show the stability and effectiveness of our approach, then discuss the real implementations in software and hardware on the NetFPGA platform. Numerical results from ns-2 simulation and experimental results from a small testbed cluster demonstrate the effectiveness of our approach in achieving high overall throughput, good fairness, smaller flow completion times (FCT) for short-lived flows, and a significant reduction in the tail of the FCT distribution.","PeriodicalId":13202,"journal":{"name":"IEEE Transactions on Cloud Computing","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2024-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10412648","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139951562","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-23DOI: 10.1109/TCC.2024.3357061
Jing Chen;Jia Chen;Kuo Guo;Renkun Hu;Tao Zou;Jun Zhu;Hongke Zhang;Jingjing Liu
In software defined network/network function virtualization (SDN/NFV)-enabled cloud environment, cloud services can be implemented as service function chains (SFCs), which consist of a series of ordered virtual network functions. However, due to fluctuations of cloud traffic and without knowledge of cloud computing network configuration, designing SFC optimization approach to obtain flexible cloud services in dynamic cloud environment is a pivotal challenge. In this paper, we propose a fault tolerance oriented SFC optimization approach based on deep reinforcement learning. We model fault tolerance oriented SFC elastic optimization problem as a Markov decision process, in which the reward is modeled as a weighted function, including minimizing energy consumption and migration cost, maximizing revenue benefit and load balancing. Then, taking binary integer programming model as constraints of quality of cloud services, we design optimization approaches for single-agent double deep Q-network (SADDQN) and multi-agent DDQN (MADDQN). Among them, MADDQN decentralizes training tasks from control plane to data plane to reduce the probability of single point of failure for the centralized controller. Experimental results show that the designed approaches have better performance. MADDQN can almost reach the upper bound of theoretical solution obtained by assuming a prior knowledge of the dynamics of cloud traffic.
{"title":"Fault Tolerance Oriented SFC Optimization in SDN/NFV-Enabled Cloud Environment Based on Deep Reinforcement Learning","authors":"Jing Chen;Jia Chen;Kuo Guo;Renkun Hu;Tao Zou;Jun Zhu;Hongke Zhang;Jingjing Liu","doi":"10.1109/TCC.2024.3357061","DOIUrl":"10.1109/TCC.2024.3357061","url":null,"abstract":"In software defined network/network function virtualization (SDN/NFV)-enabled cloud environment, cloud services can be implemented as service function chains (SFCs), which consist of a series of ordered virtual network functions. However, due to fluctuations of cloud traffic and without knowledge of cloud computing network configuration, designing SFC optimization approach to obtain flexible cloud services in dynamic cloud environment is a pivotal challenge. In this paper, we propose a fault tolerance oriented SFC optimization approach based on deep reinforcement learning. We model fault tolerance oriented SFC elastic optimization problem as a Markov decision process, in which the reward is modeled as a weighted function, including minimizing energy consumption and migration cost, maximizing revenue benefit and load balancing. Then, taking binary integer programming model as constraints of quality of cloud services, we design optimization approaches for single-agent double deep Q-network (SADDQN) and multi-agent DDQN (MADDQN). Among them, MADDQN decentralizes training tasks from control plane to data plane to reduce the probability of single point of failure for the centralized controller. Experimental results show that the designed approaches have better performance. MADDQN can almost reach the upper bound of theoretical solution obtained by assuming a prior knowledge of the dynamics of cloud traffic.","PeriodicalId":13202,"journal":{"name":"IEEE Transactions on Cloud Computing","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2024-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139956357","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-09DOI: 10.1109/TCC.2024.3351143
Lizhi Xiong;Xiao Han;Ching-Nung Yang;Yun-Qing Shi
Reversible data hiding is widely utilized for secure communication and copyright protection. Recently, to improve embedding capacity and visual quality of stego-images, some Partial Reversible Data Hiding (PRDH) schemes are proposed. But these schemes are over the plaintext domain. To protect the privacy of the cover image, Reversible Data Hiding in Encrypted Images (RDHEI) techniques are preferred. In addition, the full separability of cover image reconstruction and data restoration is also an important characteristic that cannot be achieved by most RDHEI schemes. To solve the issues, a partial and a complete Reversible Data Hiding in Shared Images with Separate Cover Image Reconstruction and Secret Extraction (RDHSI-SRE) are proposed in this paper. In the proposed schemes, the secret data is divided by Secret Sharing (SS). Then, the marked shared images are generated based on the proposed modify-and-recalculate strategy. The receiver can extract embedded data and reconstruct the image separably using k -out-of- n marked shared images. In the embedding phase of partial RDHSI-SRE (PRDHSI-SRE), the pixel values are modified according to the proposed Minimizing-Square-Errors Strategy to achieve high visual quality, and the complete RDHSI-SRE (CRDHSI-SRE) embeds data by modifying random coefficients to achieve reversibility. The experimental results and theoretical analyses demonstrate that the proposed schemes have a high embedding performance. Most importantly, the proposed schemes are fault-tolerant and completely separable.
可逆数据隐藏被广泛用于安全通信和版权保护。最近，为了提高隐去图像的嵌入能力和视觉质量，人们提出了一些部分可逆数据隐藏（PRDH）方案。但这些方案都是明文领域的。为了保护封面图像的隐私，最好采用加密图像中的可逆数据隐藏（RDHEI）技术。此外，封面图像重建和数据恢复的完全分离性也是大多数 RDHEI 方案无法实现的一个重要特征。为了解决这些问题，本文提出了一种具有分离式覆盖图像重建和秘密提取功能的共享图像中的部分和完全可逆数据隐藏（RDHSI-SRE）方案。在所提出的方案中，秘密数据通过秘密共享（Secret Sharing，SS）进行分割。然后，根据所提出的修改-计算策略生成标记共享图像。接收器可以提取嵌入的数据，并使用 n 个标记共享图像中的 k 个单独重建图像。在部分 RDHSI-SRE （PRDHSI-SRE）的嵌入阶段，根据提出的最小化平方误差策略修改像素值，以实现高视觉质量；完整 RDHSI-SRE （CRDHSI-SRE）通过修改随机系数嵌入数据，以实现可逆性。实验结果和理论分析表明，所提出的方案具有很高的嵌入性能。最重要的是，提出的方案具有容错性和完全可分性。
{"title":"Reversible Data Hiding in Shared Images With Separate Cover Image Reconstruction and Secret Extraction","authors":"Lizhi Xiong;Xiao Han;Ching-Nung Yang;Yun-Qing Shi","doi":"10.1109/TCC.2024.3351143","DOIUrl":"10.1109/TCC.2024.3351143","url":null,"abstract":"Reversible data hiding is widely utilized for secure communication and copyright protection. Recently, to improve embedding capacity and visual quality of stego-images, some Partial Reversible Data Hiding (PRDH) schemes are proposed. But these schemes are over the plaintext domain. To protect the privacy of the cover image, Reversible Data Hiding in Encrypted Images (RDHEI) techniques are preferred. In addition, the full separability of cover image reconstruction and data restoration is also an important characteristic that cannot be achieved by most RDHEI schemes. To solve the issues, a partial and a complete Reversible Data Hiding in Shared Images with Separate Cover Image Reconstruction and Secret Extraction (RDHSI-SRE) are proposed in this paper. In the proposed schemes, the secret data is divided by Secret Sharing (SS). Then, the marked shared images are generated based on the proposed modify-and-recalculate strategy. The receiver can extract embedded data and reconstruct the image separably using \u0000<italic>k</i>\u0000-out-of-\u0000<italic>n</i>\u0000 marked shared images. In the embedding phase of partial RDHSI-SRE (PRDHSI-SRE), the pixel values are modified according to the proposed Minimizing-Square-Errors Strategy to achieve high visual quality, and the complete RDHSI-SRE (CRDHSI-SRE) embeds data by modifying random coefficients to achieve reversibility. The experimental results and theoretical analyses demonstrate that the proposed schemes have a high embedding performance. Most importantly, the proposed schemes are fault-tolerant and completely separable.","PeriodicalId":13202,"journal":{"name":"IEEE Transactions on Cloud Computing","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2024-01-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139951163","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-09DOI: 10.1109/TCC.2024.3351716
Liu Liu;Zhijun Ding;Dazhao Cheng;Xiaobo Zhou
The performance of distributed ML training is largely determined by workers that generate gradients in the slowest pace, i.e., stragglers. The state-of-the-art load balancing approaches consider that each worker stores a complete dataset locally and the data fetching time can be ignored. They only consider the computation capacity of workers in equalizing the gradient computation time. However, we find that in scenarios of ML on distributed datasets, whether in edge computing or distributed data cache systems, the data fetching time is non-negligible and often becomes the primary cause of stragglers. In this paper, we present LOFT, an adaptive load balancing approach for ML upon distributed datasets at the edge. It aims to balance the time to generate gradients at each worker while ensuring the model accuracy. Specifically, LOFT features a locality-aware batching. It builds performance and optimization models upon data fetching and gradient computation time. Leveraging the models, it develops an adaptive scheme based on grid search. Furthermore, it offers Byzantine gradient aggregation upon Ring All-Reduce, which makes itself fault-tolerant under Byzantine gradients brought by a small batch size. Experiments with twelve public DNN models and four open datasets show that LOFT reduces the training time by up to 46%, while reducing the training loss by up to 67% compared to LB-BSP.
{"title":"Locality-Aware and Fault-Tolerant Batching for Machine Learning on Distributed Datasets","authors":"Liu Liu;Zhijun Ding;Dazhao Cheng;Xiaobo Zhou","doi":"10.1109/TCC.2024.3351716","DOIUrl":"10.1109/TCC.2024.3351716","url":null,"abstract":"The performance of distributed ML training is largely determined by workers that generate gradients in the slowest pace, i.e., stragglers. The state-of-the-art load balancing approaches consider that each worker stores a complete dataset locally and the data fetching time can be ignored. They only consider the computation capacity of workers in equalizing the gradient computation time. However, we find that in scenarios of ML on distributed datasets, whether in edge computing or distributed data cache systems, the data fetching time is non-negligible and often becomes the primary cause of stragglers. In this paper, we present LOFT, an adaptive load balancing approach for ML upon distributed datasets at the edge. It aims to balance the time to generate gradients at each worker while ensuring the model accuracy. Specifically, LOFT features a locality-aware batching. It builds performance and optimization models upon data fetching and gradient computation time. Leveraging the models, it develops an adaptive scheme based on grid search. Furthermore, it offers Byzantine gradient aggregation upon Ring All-Reduce, which makes itself fault-tolerant under Byzantine gradients brought by a small batch size. Experiments with twelve public DNN models and four open datasets show that LOFT reduces the training time by up to 46%, while reducing the training loss by up to 67% compared to LB-BSP.","PeriodicalId":13202,"journal":{"name":"IEEE Transactions on Cloud Computing","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2024-01-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139951172","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Deep learning (DL) has been applied in billions of mobile devices due to its astonishing performance in image, text, and audio processing. However, limited by the computing capability of mobile devices, a large amount of DL inference tasks need to be offloaded to edge or cloud servers, which makes powerful GPU servers are struggling to ensure the quality of service(QoS). To better utilize the highly parallel computing architecture of GPU to improve the QoS, we propose BatOpt, a framework that uses dynamic batch processing to strike a good balance between service latency and GPU memory usage in DL inference services. Specifically, BatOpt innovatively models the DL inference service as a $M/G(a,b)/1/N$ queue, with the consideration of stochastic task arrivals, which enables it to predict the service latency accurately in different system states. Furthermore, we propose an optimization algorithm to trade off the service latency and GPU memory usage in different system states by analyzing the queueing model. We have implemented BatOpt on Pytorch and evaluated it on an RTX 2080 GPU using real DL models. BatOpt brings up to 31x and 4.3x times performance boost in terms of service latency, compared to single-input and fixed-batch-size strategies, respectively. And BatOpt's maximum GPU memory usage is only 0.3x that of greedy-dynamic-batch-size strategy on the premise of the same service latency.
{"title":"BatOpt: Optimizing GPU-Based Deep Learning Inference Using Dynamic Batch Processing","authors":"Deyu Zhang;Yunzhen Luo;Yaobo Wang;Xiaoyan Kui;Ju Ren","doi":"10.1109/TCC.2024.3350561","DOIUrl":"10.1109/TCC.2024.3350561","url":null,"abstract":"Deep learning (DL) has been applied in billions of mobile devices due to its astonishing performance in image, text, and audio processing. However, limited by the computing capability of mobile devices, a large amount of DL inference tasks need to be offloaded to edge or cloud servers, which makes powerful GPU servers are struggling to ensure the quality of service(QoS). To better utilize the highly parallel computing architecture of GPU to improve the QoS, we propose BatOpt, a framework that uses dynamic batch processing to strike a good balance between service latency and GPU memory usage in DL inference services. Specifically, BatOpt innovatively models the DL inference service as a \u0000<inline-formula><tex-math>$M/G(a,b)/1/N$</tex-math></inline-formula>\u0000 queue, with the consideration of stochastic task arrivals, which enables it to predict the service latency accurately in different system states. Furthermore, we propose an optimization algorithm to trade off the service latency and GPU memory usage in different system states by analyzing the queueing model. We have implemented BatOpt on Pytorch and evaluated it on an RTX 2080 GPU using real DL models. BatOpt brings up to 31x and 4.3x times performance boost in terms of service latency, compared to single-input and fixed-batch-size strategies, respectively. And BatOpt's maximum GPU memory usage is only 0.3x that of greedy-dynamic-batch-size strategy on the premise of the same service latency.","PeriodicalId":13202,"journal":{"name":"IEEE Transactions on Cloud Computing","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2024-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139951519","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：