Jung-Gi Park, Un-Sook Choi, Seungwoo Kum, Jaewon Moon, Kyungyong Lee
The compute capability of edge devices is expanding owing to the wide adoption of edge computing for various application scenarios and specialized hardware explicitly developed for an edge environ-ment. A container orchestration platform, Kubernetes is widely used to maintain edge computing resources efficiently, but it suf-fers from a limited scheduling capacity. We present a design and implementation of an accelerator information extraction module to improve the scheduling capability of a standard Kubernetes imple-mentation by providing rich hardware information. Furthermore, we present a plausible advancement of the Kubernetes scheduler by considering detailed workload characteristics and attached spe-cialized accelerator hardware information.
{"title":"Accelerator-Aware Kubernetes Scheduler for DNN Tasks on Edge Computing Environment","authors":"Jung-Gi Park, Un-Sook Choi, Seungwoo Kum, Jaewon Moon, Kyungyong Lee","doi":"10.1145/3453142.3491411","DOIUrl":"https://doi.org/10.1145/3453142.3491411","url":null,"abstract":"The compute capability of edge devices is expanding owing to the wide adoption of edge computing for various application scenarios and specialized hardware explicitly developed for an edge environ-ment. A container orchestration platform, Kubernetes is widely used to maintain edge computing resources efficiently, but it suf-fers from a limited scheduling capacity. We present a design and implementation of an accelerator information extraction module to improve the scheduling capability of a standard Kubernetes imple-mentation by providing rich hardware information. Furthermore, we present a plausible advancement of the Kubernetes scheduler by considering detailed workload characteristics and attached spe-cialized accelerator hardware information.","PeriodicalId":6779,"journal":{"name":"2021 IEEE/ACM Symposium on Edge Computing (SEC)","volume":"1 1","pages":"438-440"},"PeriodicalIF":0.0,"publicationDate":"2021-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"78926741","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Martin Törngren, H. Thompson, E. Herzog, R. Inam, James Gross, G. Dán
Industry is moving towards advanced Cyber-Physical Systems (CPS), with trends in smartness, automation, connectivity and collaboration. We examine the drivers and requirements for the use of edge computing in critical industrial applications. Our purpose is to provide a better understanding of industrial needs and to initiate a discussion on what role edge computing could take, complementing current industrial and embedded systems, and the cloud. Four domains are chosen for analysis with representative use-cases; manufacturing, transportation, the energy sector and networked applications in the defense domain. We further discuss challenges, open issues and suggested directions that are needed to pave the way for the use of edge computing in industrial CPS.
{"title":"Industrial Edge-based Cyber-Physical Systems - Application Needs and Concerns for Realization","authors":"Martin Törngren, H. Thompson, E. Herzog, R. Inam, James Gross, G. Dán","doi":"10.1145/3453142.3493507","DOIUrl":"https://doi.org/10.1145/3453142.3493507","url":null,"abstract":"Industry is moving towards advanced Cyber-Physical Systems (CPS), with trends in smartness, automation, connectivity and collaboration. We examine the drivers and requirements for the use of edge computing in critical industrial applications. Our purpose is to provide a better understanding of industrial needs and to initiate a discussion on what role edge computing could take, complementing current industrial and embedded systems, and the cloud. Four domains are chosen for analysis with representative use-cases; manufacturing, transportation, the energy sector and networked applications in the defense domain. We further discuss challenges, open issues and suggested directions that are needed to pave the way for the use of edge computing in industrial CPS.","PeriodicalId":6779,"journal":{"name":"2021 IEEE/ACM Symposium on Edge Computing (SEC)","volume":"23 1","pages":"409-415"},"PeriodicalIF":0.0,"publicationDate":"2021-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87258879","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Action recognition is a key algorithmic part of emerging on-the-edge smart video surveillance and security systems. Skeleton-based action recognition is an attractive approach which, instead of using RGB pixel data, relies on human pose information to classify appropriate actions. However, existing algorithms often assume ideal conditions that are not representative of real-world limitations, such as noisy input, latency requirements, and edge resource constraints. To address the limitations of existing approaches, this paper presents Real-World Graph Convolution Networks (RW-GCNs), an architecture-level solution for meeting the domain constraints of Real World Skeleton-based Action Recognition. Inspired by the presence of feedback connections in the human visual cortex, RW-GCNs leverage attentive feedback augmentation on existing near state-of-the-art (SotA) Spatial-Temporal Graph Convolution Net-works (ST-GCNs). The ST-GCNs' design choices are derived from information theory-centric principles to address both the spatial and temporal noise typically encountered in end-to-end real-time and on-the-edge smart video systems. Our results demonstrate RW-GCNs' ability to serve these applications by achieving a new SotA accuracy on the NTU-RGB-D-120 dataset at 94.1%, and achieving 32× less latency than baseline ST-GCN applications while still achieving 90.4% accuracy on the Northwestern UCLA dataset in the presence of spatial keypoint noise. RW-GCNs further show system scalability by running on the 10× cost effective NVIDIA Jetson Nano (as opposed to NVIDIA Xavier NX), while still main-taining a respectful range of throughput (15.6 to 5.5 Actions per Second) on the resource constrained device. The code is available here: https://github.com/TeCSAR-UNCC/RW-GCN.
{"title":"Real-World Graph Convolution Networks (RW-GCNs) for Action Recognition in Smart Video Surveillance","authors":"Justin Sanchez, Christopher Neff, H. Tabkhi","doi":"10.1145/3453142.3491293","DOIUrl":"https://doi.org/10.1145/3453142.3491293","url":null,"abstract":"Action recognition is a key algorithmic part of emerging on-the-edge smart video surveillance and security systems. Skeleton-based action recognition is an attractive approach which, instead of using RGB pixel data, relies on human pose information to classify appropriate actions. However, existing algorithms often assume ideal conditions that are not representative of real-world limitations, such as noisy input, latency requirements, and edge resource constraints. To address the limitations of existing approaches, this paper presents Real-World Graph Convolution Networks (RW-GCNs), an architecture-level solution for meeting the domain constraints of Real World Skeleton-based Action Recognition. Inspired by the presence of feedback connections in the human visual cortex, RW-GCNs leverage attentive feedback augmentation on existing near state-of-the-art (SotA) Spatial-Temporal Graph Convolution Net-works (ST-GCNs). The ST-GCNs' design choices are derived from information theory-centric principles to address both the spatial and temporal noise typically encountered in end-to-end real-time and on-the-edge smart video systems. Our results demonstrate RW-GCNs' ability to serve these applications by achieving a new SotA accuracy on the NTU-RGB-D-120 dataset at 94.1%, and achieving 32× less latency than baseline ST-GCN applications while still achieving 90.4% accuracy on the Northwestern UCLA dataset in the presence of spatial keypoint noise. RW-GCNs further show system scalability by running on the 10× cost effective NVIDIA Jetson Nano (as opposed to NVIDIA Xavier NX), while still main-taining a respectful range of throughput (15.6 to 5.5 Actions per Second) on the resource constrained device. The code is available here: https://github.com/TeCSAR-UNCC/RW-GCN.","PeriodicalId":6779,"journal":{"name":"2021 IEEE/ACM Symposium on Edge Computing (SEC)","volume":"33 1","pages":"121-134"},"PeriodicalIF":0.0,"publicationDate":"2021-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"86422989","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Modern EdgeAI inference systems still have many cruciallimi-tations. In this paper, we holistically consider implications and optimizations of EdgeAI inference systems for object detection applications in efficiency and accuracy. We summarize three in-trinsic limitations of current-generation EdgeAI inference systems based on our observations (i.e., less compute capabilities, restrictions of operations, and accuracy loss due to numerical precision). Then we propose three approaches to improve end-to-end performance and prediction accuracy: 1) Utilizing parallel computing designs and methods to solve computational bottlenecks; 2) Ap-plying domain-specific optimizations to mostly eliminate accuracy loss; 3) Using higher-quality input data to saturate the processors and accelerators. We also provide five recommendations for end-to-end EdgeAI solution deployments, which are usually neglected by EdgeAI users. In particular, we deploy and optimize two real object detection applications (2D and 3D) on two EdgeAI inference systems (NovuTensor and Nvidia Xavier) with widely used datasets (i.e., MS-COCO, PASCAL-VOC, and KITTI). The results show that runtime performance can be accelerated by up to 2X on NovuTen-sor and the mean average precision (mAP) can be increased by 46% through applying our proposed methods.
{"title":"Characterizing and Accelerating End-to-End EdgeAI Inference Systems for Object Detection Applications","authors":"Yujie Hui, J. Lien, Xiaoyi Lu","doi":"10.1145/3453142.3491294","DOIUrl":"https://doi.org/10.1145/3453142.3491294","url":null,"abstract":"Modern EdgeAI inference systems still have many cruciallimi-tations. In this paper, we holistically consider implications and optimizations of EdgeAI inference systems for object detection applications in efficiency and accuracy. We summarize three in-trinsic limitations of current-generation EdgeAI inference systems based on our observations (i.e., less compute capabilities, restrictions of operations, and accuracy loss due to numerical precision). Then we propose three approaches to improve end-to-end performance and prediction accuracy: 1) Utilizing parallel computing designs and methods to solve computational bottlenecks; 2) Ap-plying domain-specific optimizations to mostly eliminate accuracy loss; 3) Using higher-quality input data to saturate the processors and accelerators. We also provide five recommendations for end-to-end EdgeAI solution deployments, which are usually neglected by EdgeAI users. In particular, we deploy and optimize two real object detection applications (2D and 3D) on two EdgeAI inference systems (NovuTensor and Nvidia Xavier) with widely used datasets (i.e., MS-COCO, PASCAL-VOC, and KITTI). The results show that runtime performance can be accelerated by up to 2X on NovuTen-sor and the mean average precision (mAP) can be increased by 46% through applying our proposed methods.","PeriodicalId":6779,"journal":{"name":"2021 IEEE/ACM Symposium on Edge Computing (SEC)","volume":"205 1","pages":"01-12"},"PeriodicalIF":0.0,"publicationDate":"2021-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"75865909","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We present CAPLets, an authorization mechanism that extends capability based security to support fine grained access control for multi-scale (sensors, edge, cloud) IoT deployments. To enable this, CAPLets uses a strong cryptographic construction to provide integrity while preserving computational efficiency for resource constrained systems. Moreover, CAPLets augments capabilities with dynamic, user defined constraints to describe arbitrary access control policies. We introduce an application specific, turing complete virtual machine, CapVM, alongside with eBPF and Wasm, to describe constraints. We show that CAPLets is able to express permissions and requirements at a fine grain, facilitating construction of non-trivial access control policies. We empirically evaluate the efficiency and flexibility of CAPLets abstractions using resource constrained devices and end-to-end IoT deployments, and compare it against related mechanisms in wide use today. Our empirical results show that CAPLets is an order of magnitude faster and more energy efficient than current IoT authorization systems.
{"title":"CAPLets: Resource Aware, Capability-Based Access Control for IoT","authors":"F. Bakir, C. Krintz, R. Wolski","doi":"10.1145/3453142.3491289","DOIUrl":"https://doi.org/10.1145/3453142.3491289","url":null,"abstract":"We present CAPLets, an authorization mechanism that extends capability based security to support fine grained access control for multi-scale (sensors, edge, cloud) IoT deployments. To enable this, CAPLets uses a strong cryptographic construction to provide integrity while preserving computational efficiency for resource constrained systems. Moreover, CAPLets augments capabilities with dynamic, user defined constraints to describe arbitrary access control policies. We introduce an application specific, turing complete virtual machine, CapVM, alongside with eBPF and Wasm, to describe constraints. We show that CAPLets is able to express permissions and requirements at a fine grain, facilitating construction of non-trivial access control policies. We empirically evaluate the efficiency and flexibility of CAPLets abstractions using resource constrained devices and end-to-end IoT deployments, and compare it against related mechanisms in wide use today. Our empirical results show that CAPLets is an order of magnitude faster and more energy efficient than current IoT authorization systems.","PeriodicalId":6779,"journal":{"name":"2021 IEEE/ACM Symposium on Edge Computing (SEC)","volume":"8 1","pages":"106-120"},"PeriodicalIF":0.0,"publicationDate":"2021-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"78456548","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
A. Liu, O.M.K. Law, Jeremiah Liao, Jeffrey Y.C. Chen, Andy Jia En Hsieh, C. Hsieh
Due to the surging of mobile data, edge AI is developed to address the cloud limitations, real-time processing, data latency, network bandwidth, and power dissipation. Kneron successfully implements the traffic safety system using the smart edge AI architecture, which applies the smart gateway to bridge the gap between cloud and edge AI, it also establishes the open platform for further integration. New architectures offer additional security and privacy protection through the blockchain approach.
{"title":"Traffic Safety System Edge AI Computing","authors":"A. Liu, O.M.K. Law, Jeremiah Liao, Jeffrey Y.C. Chen, Andy Jia En Hsieh, C. Hsieh","doi":"10.1145/3453142.3491410","DOIUrl":"https://doi.org/10.1145/3453142.3491410","url":null,"abstract":"Due to the surging of mobile data, edge AI is developed to address the cloud limitations, real-time processing, data latency, network bandwidth, and power dissipation. Kneron successfully implements the traffic safety system using the smart edge AI architecture, which applies the smart gateway to bridge the gap between cloud and edge AI, it also establishes the open platform for further integration. New architectures offer additional security and privacy protection through the blockchain approach.","PeriodicalId":6779,"journal":{"name":"2021 IEEE/ACM Symposium on Edge Computing (SEC)","volume":"14 1","pages":"01-02"},"PeriodicalIF":0.0,"publicationDate":"2021-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"79931147","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Murugan Sankaradas, Kunal Rao, Ravi Rajendran, Amit Redkar, S. Chakradhar
Edge computing and 5G have made it possible to perform analytics closer to the source of data and achieve super-low latency response times, which isn't possible with centralized cloud deployment. In this paper, we present a novel fever screening system, which uses edge machine learning techniques and leverages private 5G to accurately identify and screen individuals with fever in real-time. Particularly, we present deep-learning based novel techniques for fusion and alignment of cross-spectral visual and thermal data streams at the edge. Our novel Cross-Spectral Generative Adversarial Network (CS-GAN) synthesizes visual images that have the key, representative object level features required to uniquely associate objects across visual and thermal spectrum. Two key features of CS-GAN are a novel, feature-preserving loss function that results in high-quality pairing of corresponding cross-spectral objects, and dual bottleneck residual layers with skip connections (a new, network enhancement) to not only accelerate real-time inference, but to also speed up convergence during model training at the edge. To the best of our knowledge, this is the first technique that leverages 5G networks and limited edge resources to enable real-time feature-level association of objects in visual and thermal streams (30 ms per full HD frame on an Intel Core i7-8650 4-core, 1.9GHz mobile processor). To the best of our knowledge, this is also the first system to achieve real-time operation, which has enabled fever screening of employees and guests in arenas, theme parks, airports and other critical facilities. By leveraging edge computing and 5G, our fever screening system is able to achieve 98.5% accuracy and is able to process ∼ 5X more people when compared to a centralized cloud deployment.
{"title":"Edge-based fever screening system over private 5G","authors":"Murugan Sankaradas, Kunal Rao, Ravi Rajendran, Amit Redkar, S. Chakradhar","doi":"10.1145/3453142.3493516","DOIUrl":"https://doi.org/10.1145/3453142.3493516","url":null,"abstract":"Edge computing and 5G have made it possible to perform analytics closer to the source of data and achieve super-low latency response times, which isn't possible with centralized cloud deployment. In this paper, we present a novel fever screening system, which uses edge machine learning techniques and leverages private 5G to accurately identify and screen individuals with fever in real-time. Particularly, we present deep-learning based novel techniques for fusion and alignment of cross-spectral visual and thermal data streams at the edge. Our novel Cross-Spectral Generative Adversarial Network (CS-GAN) synthesizes visual images that have the key, representative object level features required to uniquely associate objects across visual and thermal spectrum. Two key features of CS-GAN are a novel, feature-preserving loss function that results in high-quality pairing of corresponding cross-spectral objects, and dual bottleneck residual layers with skip connections (a new, network enhancement) to not only accelerate real-time inference, but to also speed up convergence during model training at the edge. To the best of our knowledge, this is the first technique that leverages 5G networks and limited edge resources to enable real-time feature-level association of objects in visual and thermal streams (30 ms per full HD frame on an Intel Core i7-8650 4-core, 1.9GHz mobile processor). To the best of our knowledge, this is also the first system to achieve real-time operation, which has enabled fever screening of employees and guests in arenas, theme parks, airports and other critical facilities. By leveraging edge computing and 5G, our fever screening system is able to achieve 98.5% accuracy and is able to process ∼ 5X more people when compared to a centralized cloud deployment.","PeriodicalId":6779,"journal":{"name":"2021 IEEE/ACM Symposium on Edge Computing (SEC)","volume":"14 1","pages":"386-391"},"PeriodicalIF":0.0,"publicationDate":"2021-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87781505","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Kumseok Jung, Julien Gascon-Samson, K. Pattabiraman
Edge computing application developers often need to employ a combination of software tools in order to deal with the challenges of heterogeneity and network dynamism. As a result, developers write extra code irrelevant to the core application logic, to provide interoperability between interacting tools. Existing software frameworks offer programming models and cloud-hosted services to ease the overall development process. However, the framework-specific APIs exacerbate the technology fragmentation problem, requiring developers to write more glue code between competing frameworks. In this paper, we present a middleware called OneOS, which provides a distributed computing environment through the standard POSIX API. OneOS maintains a global view of the computer network, presenting the same file system and process space to any user application running in the network. OneOS intercepts POSIX API calls and transparently handles the interaction with the corresponding I/O resource in the network. Using the OneOS Domain-Specific Language (DSL), users can distribute a legacy POSIX pipeline over the network. We evaluate the performance of OneOS against an open-source IoT Platform, ThingsJS, using an IoT stream processing benchmark suite, and a distributed video processing application. OneOS executes the programs about 3x faster than ThingsJS, and reduces the code size by about 25%.
{"title":"OneOS: Middleware for Running Edge Computing Applications as Distributed POSIX Pipelines","authors":"Kumseok Jung, Julien Gascon-Samson, K. Pattabiraman","doi":"10.1145/3453142.3493505","DOIUrl":"https://doi.org/10.1145/3453142.3493505","url":null,"abstract":"Edge computing application developers often need to employ a combination of software tools in order to deal with the challenges of heterogeneity and network dynamism. As a result, developers write extra code irrelevant to the core application logic, to provide interoperability between interacting tools. Existing software frameworks offer programming models and cloud-hosted services to ease the overall development process. However, the framework-specific APIs exacerbate the technology fragmentation problem, requiring developers to write more glue code between competing frameworks. In this paper, we present a middleware called OneOS, which provides a distributed computing environment through the standard POSIX API. OneOS maintains a global view of the computer network, presenting the same file system and process space to any user application running in the network. OneOS intercepts POSIX API calls and transparently handles the interaction with the corresponding I/O resource in the network. Using the OneOS Domain-Specific Language (DSL), users can distribute a legacy POSIX pipeline over the network. We evaluate the performance of OneOS against an open-source IoT Platform, ThingsJS, using an IoT stream processing benchmark suite, and a distributed video processing application. OneOS executes the programs about 3x faster than ThingsJS, and reduces the code size by about 25%.","PeriodicalId":6779,"journal":{"name":"2021 IEEE/ACM Symposium on Edge Computing (SEC)","volume":"22 1","pages":"242-256"},"PeriodicalIF":0.0,"publicationDate":"2021-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"86714715","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In order to solve security and privacy issues of centralized cloud services, the edge computing network is introduced, where computing and storage resources are distributed to the edge of the network. However, native edge computing is subject to the limited performance of edge devices, which causes challenges in data authorization, data encryption, user privacy, and other fields. Blockchain is currently the hottest technology for distributed networks. It solves the consistent issue of distributed data and is used in many areas, such as cryptocurrency, smart grid, and the Internet of Things. Our work discussed the security and privacy challenges of edge computing networks. From the perspectives of data authorization, encryption, and user privacy, we analyze the solutions brought by blockchain technology to edge computing networks. In this work, we deeply present the benefits from the integration of the edge computing network and blockchain technology, which effectively controls the data authorization and data encryption of the edge network and enhances the architecture's scalability under the premise of ensuring security and privacy. Finally, we investigate challenges on storage, workload, and latency for future research in this field.
{"title":"How BlockChain Can Help Enhance The Security And Privacy in Edge Computing?","authors":"Jinyue Song, Tianbo Gu, P. Mohapatra","doi":"10.1145/3453142.3493513","DOIUrl":"https://doi.org/10.1145/3453142.3493513","url":null,"abstract":"In order to solve security and privacy issues of centralized cloud services, the edge computing network is introduced, where computing and storage resources are distributed to the edge of the network. However, native edge computing is subject to the limited performance of edge devices, which causes challenges in data authorization, data encryption, user privacy, and other fields. Blockchain is currently the hottest technology for distributed networks. It solves the consistent issue of distributed data and is used in many areas, such as cryptocurrency, smart grid, and the Internet of Things. Our work discussed the security and privacy challenges of edge computing networks. From the perspectives of data authorization, encryption, and user privacy, we analyze the solutions brought by blockchain technology to edge computing networks. In this work, we deeply present the benefits from the integration of the edge computing network and blockchain technology, which effectively controls the data authorization and data encryption of the edge network and enhances the architecture's scalability under the premise of ensuring security and privacy. Finally, we investigate challenges on storage, workload, and latency for future research in this field.","PeriodicalId":6779,"journal":{"name":"2021 IEEE/ACM Symposium on Edge Computing (SEC)","volume":"22 1","pages":"448-453"},"PeriodicalIF":0.0,"publicationDate":"2021-10-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"88192680","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Training on the Edge enables neural networks to learn continu-ously from new data after deployment on memory-constrained edge devices. Previous work is mostly concerned with reducing the number of model parameters which is only beneficial for in-ference. However, memory footprint from activations is the main bottleneck for training on the edge. Existing incremental training methods fine-tune the last few layers sacrificing accuracy gains from re-training the whole model. In this work, we investigate the memory footprint of training deep learning models, and use our observations to propose BitTrain. In BitTrain, we exploit activation sparsity and propose a novel bitmap compression technique that reduces the memory footprint during training. We save the activations in our proposed bitmap compression format during the forward pass of the training, and restore them during the backward pass for the optimizer computations. The proposed method can be integrated seamlessly in the computation graph of modern deep learning frameworks. Our implementation is safe by construction, and has no negative impact on the accuracy of model training. Experimental results show up to 34% reduction in the memory footprint at a sparsity level of 50%. Further pruning during training results in more than 70% sparsity, which can lead to up to 56% re-duction in memory footprint. BitTrain advances the efforts towards bringing more machine learning capabilities to edge devices. Our source code is available at https://github.com/scale-lab/BitTrain.
{"title":"Sparse Bitmap Compression for Memory-Efficient Training on the Edge","authors":"Abdelrahman Hosny, Marina Neseem, S. Reda","doi":"10.1145/3453142.3491290","DOIUrl":"https://doi.org/10.1145/3453142.3491290","url":null,"abstract":"Training on the Edge enables neural networks to learn continu-ously from new data after deployment on memory-constrained edge devices. Previous work is mostly concerned with reducing the number of model parameters which is only beneficial for in-ference. However, memory footprint from activations is the main bottleneck for training on the edge. Existing incremental training methods fine-tune the last few layers sacrificing accuracy gains from re-training the whole model. In this work, we investigate the memory footprint of training deep learning models, and use our observations to propose BitTrain. In BitTrain, we exploit activation sparsity and propose a novel bitmap compression technique that reduces the memory footprint during training. We save the activations in our proposed bitmap compression format during the forward pass of the training, and restore them during the backward pass for the optimizer computations. The proposed method can be integrated seamlessly in the computation graph of modern deep learning frameworks. Our implementation is safe by construction, and has no negative impact on the accuracy of model training. Experimental results show up to 34% reduction in the memory footprint at a sparsity level of 50%. Further pruning during training results in more than 70% sparsity, which can lead to up to 56% re-duction in memory footprint. BitTrain advances the efforts towards bringing more machine learning capabilities to edge devices. Our source code is available at https://github.com/scale-lab/BitTrain.","PeriodicalId":6779,"journal":{"name":"2021 IEEE/ACM Symposium on Edge Computing (SEC)","volume":"14 1","pages":"14-25"},"PeriodicalIF":0.0,"publicationDate":"2021-10-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73410987","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","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：