Convergence of Edge Computing and Deep Learning: A Comprehensive Survey

Probably not a substantial claim: may not pass S3 (load-bearing).

Various deep learning-based intelligent services and applications have changed many aspects of people's lives due to DL advantages in computer vision and natural language processing.

“As a typical and more widely used new form of applications [9], various deep learning-based intelligent services and applications have changed many aspects of people's lives due to the great advantages of Deep Learning (DL) in the fields of Computer Vision (CV) and Natural Language Processing (NLP) [10].”

• cites [9] sourcePartly supported

  The Introduction (p. 1) describes potential applications of fast object detection in driving, assistive devices, and robotics, supporting the computer-vision portion of the claim. It does not address natural language processing or establish that deep-learning applications have already changed many aspects of people's lives. (LLM, conf=0.86)

  Experimental — accuracy not yet measured
• cites [10] sourcePartly supported

  The Introduction (p. 4) reports deep learning outperforming other methods in numerous applications, and §5.22 (p. 25) describes advances in speech recognition and machine translation. Section 5.21 (p. 24) describes computer-vision applications in medical diagnosis, but frames their wider benefits as possibilities. These passages support the claimed advantages in vision and language-related tasks, not the assertion that applications have already changed many aspects of people’s lives. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

May combine two claims.

With network slicing, customized and optimized resources can be provided for Edge DL services, helping reduce latency caused by access networks and support dense access to these services.

“With network slicing, customized and optimized resources can be provided for Edge DL services, which can help reduce latency caused by access networks and support dense access to these services [77].”

• cites [77] sourceCould not verify

  The Introduction (p. 2) discusses compressing neural networks for mobile devices, and Future Work (p. 12) discusses customized hardware for those compressed models. Neither addresses network slicing, access-network latency, or dense access to Edge DL services; searches for those terms and close paraphrases found no relevant passage. (LLM, conf=0.78)

  Experimental — accuracy not yet measured

May combine two claims.

For computation-intensive tasks, segmenting and dispatching them to the end, edge and cloud reduces execution delay while ensuring accuracy.

“For a computation-intensive task, it will be reasonably segmented and dispatched separately to the end, edge and cloud for execution, reducing the execution delay of the task while ensuring the accuracy of the results [12], [49], [50].”

• cites [12] sourceNot checked — no full text of the source
• cites [49] sourcePartly supported

  Section 2.3 describes partitioning a neural network between an edge device and the cloud, not dispatching a task across separate end, edge, and cloud tiers. Section 3.1 reports little accuracy loss and faster inference for some tested networks, but its table shows speed-ups below 1× for others. The source therefore does not support a general claim that this arrangement reduces execution delay. (LLM, conf=0.88)

  Experimental — accuracy not yet measured
• cites [50] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

EdgeEye presents an edge service framework that realizes a high-level abstraction of real-time video analytic functions based on DL.

“Besides, in order to optimize performance and efficiency, [108] presents an edge service framework, i.e., EdgeEye, which realizes a high-level abstraction of real-time video analytic functions based on DL.”

• cites [108] sourceSupported

May combine two claims.

NestDNN's multi-capacity model provides variable resource-accuracy trade-offs with a compact memory footprint, ensuring efficient multi-tenant DL at the end level.

“In this way, the multi-capacity model provides variable resourceaccuracy trade-offs with a compact memory footprint, hence ensuring efficient multi-tenant DL at the end level.”

• cites [105] sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

DeepIns partitions the DL model used for inspection and deploys them on the end, edge and cloud layer separately for improving inspection efficiency.

“The main idea of this system is partitioning the DL model, used for inspection, and deploying them on the end, edge and cloud layer separately for improving the inspection efficiency.”

• cites [113] sourcePartly supported

  The Deep Model section (p. 4) says cameras capture images, fog nodes perform lower-level inference, and cloud servers complete inference when needed. The Offloading Strategy section (p. 6) describes allocating model layers to fog devices and cloud servers to improve efficiency. It does not describe deploying a separate part of the model on an end-device layer, so the claimed three-layer deployment is not supported. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

Intelligent manufacturing should first address response latency, risk control, and privacy protection, and hence requires DL and edge computing.

“In order to follow these principles, intelligent manufacturing should first address response latency, risk control, and privacy protection, and hence requires DL and edge computing.”

• cites [113] sourcePartly supported

  The Introduction (p. 2) describes response latency as a concern and proposes deep learning with fog computing for real-time manufacturing inspection. It does not establish that intelligent manufacturing generally requires these technologies, or identify risk control and privacy protection as reasons for adopting them. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

May combine two claims.

DeepIns uses DL and edge computing to guarantee performance and process delay respectively.

“For autonomous manufacturing inspection, DeepIns [113] uses DL and edge computing to guarantee performance and process delay respectively.”

• cites [113] sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

MAUI can decompose the whole program at runtime instead of manually partitioning before program deployment.

“this solution can decompose the whole program at runtime instead of manually partitioning of programmers before program deploying.”

• cites [178] sourceCould not verify

  The Introduction (p. 2) and Federated Dropout section (p. 4) discuss reducing federated-learning model size and client computation, not decomposing a program at runtime. Searches for MAUI, runtime program partitioning, and pre-deployment manual partitioning found no passage addressing the claim. (LLM, conf=0.76)

  Experimental — accuracy not yet measured

May combine two claims.

RAN controllers deployed at edge nodes collect data and run DL services, while a cloud network controller orchestrates their operations, providing DL services to users at network edge.

“The RAN controllers deployed at edge nodes are introduced in [180] to collect the data and run DL services, while the network controller, placed in the cloud, orchestrates the operations of the RAN controllers.In this manner, after running and feeding analytics and extract relevant metrics to DL models, these controllers can provide DL services to the users at the network edge.”

• cites [180] Polese, 2018 sourcePartly supported

  The Contributions section (p. 2) describes edge-deployed RAN controllers collecting data and running an ML infrastructure, with a cloud controller orchestrating them and the architecture providing new services to users. The source says machine learning, however, not specifically deep learning (DL); its listed techniques are regressors. The claim’s DL-specific description is therefore unsupported. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing).

[222] expands communication by 1.73×-1.98×.

“Communication expansion: 1.73×-1.98×”

• cites [222] sourceCould not verify

  Sections 3.3–3.4 (pp. 4–5) report computational cost and parameter counts, not communication expansion. Searches for communication overhead and the 1.73×–1.98× range found no passage addressing the claim. (LLM, conf=0.76)

  Experimental — accuracy not yet measured

May combine two claims.

q-FFL minimizes an aggregate re-weighted loss parameterized by q, so that devices with higher loss are given higher relative weight, thus encouraging less variance (i.e., more fairness) in the accuracy distribution.

“q-FFL minimizes an aggregate re-weighted loss parameterized by q, so that devices with higher loss are given higher relative weight, thus encouraging less variance (i.e., more fairness) in the accuracy distribution.”

• cites [221] Li, 2019 sourceSupported
• cites [226] sourceNot checked — no full text of the source

May combine two claims.

[205] reduces downlink by 14×, uplink by 28×, and local computation by 1.7×.

“Downlink reduction: 14×; Uplink reduction: 28×; Local computation reduction: 1.7×”

• cites [205] Caldas, 2018 sourceCould not verify

  The Introduction (p. 1) discusses reducing storage and computational costs, and Quantization and Binarization (p. 2) describes model compression. Neither passage reports downlink, uplink, and local-computation reductions of 14×, 28×, and 1.7×. Searches for those terms and figures found no passage addressing the claimed result. (LLM, conf=0.73)

  Experimental — accuracy not yet measured

May combine two claims.

FederatedAveraging is robust to unbalanced and non-IID data and reduces communication rounds needed to train a DL model.

“Experimental results corroborate Federate-dAveraging is robust to unbalanced and non-IID data and can facilitate the training process, viz., reducing the rounds of communication needed to train a DL model.”

• cites [198] McMahan, Communication-efficient learning of deep networks from decentralized data sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

[199] improves scalability up to 1.5e6 clients.

“Scalability improvement: up to 1.5e6 clients”

• cites [199] Bonawitz, 2019 sourcePartly supported

  The Operational Profile (p. 9) supports the general scalability claim: the system handled approximately 10 million daily active devices. It does not report 1.5 million clients as a scalability limit or measured result; it separately reports up to 10,000 devices participating simultaneously. (LLM, conf=0.85)

  Experimental — accuracy not yet measured

May combine two claims.

[212] improves Top 1 accuracy by 5.59%-5.89% and reduces communication traffic by 92%.

“Top 1 accuracy imrpovement: 5.59%-5.89%;Communication traffic reduction: 92%”

• cites [212] Duan, 2019 sourcePartly supported

  The Conclusion (p. 8) reports top-1 accuracy gains of 5.59% on imbalanced EMNIST and 5.89% on imbalanced CINIC-10 versus FedAvg. The Overhead section (p. 8), however, reports an 81.7% reduction in communication cost for its measured comparison, not 92%. (LLM, conf=0.91)

  Experimental — accuracy not yet measured

May combine two claims.

Adaptively minimizing q-FFL avoids the burden of handcrafting fairness constraints, and can adjust the goal according to the required fairness dynamically, achieving the effect of reducing the variance of accuracy distribution among participated devices.

“Adaptively minimizing q-FFL avoids the burden of handcrafting fairness constraints, and can adjust the goal according to the required fairness dynamically, achieving the effect of reducing the variance of accuracy distribution among partici-pated devices.”

• cites [221] Li, 2019 sourceSupported

May combine two claims.

DGC exchanges only important gradients larger than a heuristically given threshold, accumulates the rest locally until they exceed the threshold, and codes/compresses gradients to save communication cost.

“First, DGC stipulates that only important gradients are exchanged, i.e., only gradients larger than a heuristically given threshold are transmitted.In order to avoid the information losing, the rest of the gradients are accumulated locally until they exceed the threshold.To be noted, gradients whether being immediately transmitted or accumulated for later exchanging will be coded and compressed, a”

• cites [191] Lin, 2017 sourceSupported

• AssumptionRests on an assumption: Most of the gradient exchanges are redundant.

May combine two claims.

Most gradient exchanges are redundant, so updated gradients can be compressed to cut communication cost while preserving training accuracy.

“Most of the gradient exchanges are redundant, and hence updated gradients can be compressed to cut down the communication cost while preserving the training accuracy (such as DGC in [191]).”

• cites [191] Lin, 2017 sourceSupported

• AssumptionRests on an assumption: Most of the gradient exchanges are redundant.

May combine two claims.

FL allows edge devices to train local DL models with their own data and upload only updated model, without uploading data for central cloud training.

“Without requiring uploading data for central cloud training, FL [198], [199] can allow edge devices to train their local DL models with their own collected data and upload only the updated model instead.”

• cites [198] McMahan, Communication-efficient learning of deep networks from decentralized data sourceSupported
• cites [199] Bonawitz, 2019 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

[198] reduces communication rounds by 10-100×.

“Communication rounds reduction: 10-100×”

• cites [198] McMahan, Communication-efficient learning of deep networks from decentralized data sourceSupported

May combine two claims.

By using cross-modal transfer in the learning of edge devices across different sensing modalities, required labeled data and the training process can be largely reduced and accelerated, respectively.

“By using the cross-modal transfer in the learning of edge devices across different sensing modalities, required labeled data and the training process can be largely reduced and accelerated, respectively.”

• cites [284] sourceSupported

May combine two claims.

Existing FL methods focus on synchronous training, and can only process hundreds of devices in parallel.

“Existing FL methods [198], [199] focus on synchronous training, and can only process hundreds of devices in parallel.”

• cites [198] McMahan, Communication-efficient learning of deep networks from decentralized data sourcePartly supported

  The Introduction (p. 2) says this paper assumes a synchronous update scheme, supporting that characterization of its method. It does not say existing federated-learning methods can only process hundreds of devices in parallel; the 100-client figure in the Experimental Results (p. 5) describes one experimental setup, not a capacity limit. (LLM, conf=0.87)

  Experimental — accuracy not yet measured
• cites [199] Bonawitz, 2019 sourcePartly supported

  The Introduction (p. 1) says the authors chose to support synchronous rounds, and Future Work (p. 10) says current FL algorithms can efficiently utilize only hundreds of devices in parallel. The claim drops the efficiency qualifier: the Operational Profile (p. 9) reports up to 10,000 devices participating simultaneously, so the paper does not establish a hard limit of hundreds that can be processed. (LLM, conf=0.87)

  Experimental — accuracy not yet measured

May combine two claims.

The application of vanilla FL is using RNN for next-word-prediction, in which the training data can be obtained along with users' daily inputs.

“The application of vanilla FL is using RNN for next-word-prediction [199], in which the training data can be obtained along with users' daily inputs.”

• cites [199] Bonawitz, 2019 sourceSupported

May combine two claims.

To solve this kind of task offloading problem is NP-hard, since at least combination optimization of communication and computing resources along with the contention of edge devices is required.

“To solve this kind of task offloading problem is NP-hard [244], since at least combination optimization of communication and computing resources along with the contention of edge devices is required.”

• cites [244] sourcePartly supported

  The Introduction (p. 2) says finding a centralized optimal multi-user offloading solution is NP-hard in a multi-channel wireless interference environment, and that the model accounts for communication and computation. But it treats wireless contention as a separate extension; the cited passages do not establish that contention is required for, or explains, the NP-hardness claim. Related Work (p. 12) also limits the result to the generalized multi-channel problem, noting that the single-channel case is not NP-hard. (LLM, conf=0.85)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

The data required for DL might carry a lot of private information, and privacy issues are critical to areas such as smart home and cities.

“Privacy: the data required for DL might carry a lot of private information, and privacy issues are critical to areas such as smart home and cities.”

May combine two claims.

A large number of computation tasks need to be delivered to the cloud for processing, posing serious challenges on network capacity and cloud computing infrastructure.

“a large number of computation tasks need to be delivered to the cloud for processing [3], which undoubtedly poses serious challenges on network capacity and the computing power of cloud computing infrastructures;”

• cites [3] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S4 (based).

The hierarchical computing architecture provides more reliable DL computation.

“3) the hierarchical computing architecture provides more reliable DL computation;”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Most cloud computing applications rely on wireless communications and backbone networks for connecting users to services, but many industrial scenarios require intelligent services to be highly reliable even when network connections are lost.

“Reliability: most cloud computing applications relies on wireless communications and backbone networks for connecting users to services, but for many industrial scenarios, intelligent services must be highly reliable, even when network connections are lost;”

May combine two claims.

Training and inference of DL models in the cloud requires devices or users to transmit massive amounts of data to the cloud, consuming a large amount of network bandwidth.

“Cost: training and inference of DL models in the cloud requires devices or users to transmit massive amounts of data to the cloud, thus consuming a large amount of network bandwidth;”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.

With richer data and application scenarios, edge computing can promote the pervasive application of DL and realize the prospect of providing AI for every person and every organization at everywhere.

“4) with richer data and application scenarios, edge computing can promote the pervasive application of DL and realize the prospect of”

• cites [13] 2019Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

The delay to access cloud services is generally not guaranteed and might not be short enough to satisfy the requirements of many time-critical applications such as cooperative autonomous driving.

“Latency: the delay to access cloud services is generally not guaranteed and might not be short enough to satisfy the requirements of many time-critical applications such as cooperative autonomous driving [11];”

• cites [11] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Edge computing is gradually being combined with artificial intelligence, benefiting each other in terms of the realization of edge intelligence and intelligent edge.

“In fact, edge computing is gradually being combined with Artificial Intelligence (AI), benefiting each other in terms of the realization of edge intelligence and intelligent edge as depicted in Fig. 1.”

Classified as probably not a substantial claim.

Since the edge is closer to users than the cloud, edge computing is expected to solve many of the issues of cloud-based intelligent services.

“Since the edge is closer to users than the cloud, edge computing is expected to solve many of these issues.”

• AssumptionRests on an assumption: The edge is closer to users than the cloud.

Classified as probably not a substantial claim. May combine two claims.

Edge intelligence pushes DL computations from the cloud to the edge as much as possible, enabling various distributed, low-latency and reliable intelligent services.

“on one hand, edge intelligence is expected to push DL computations from the cloud to the edge as much as possible, thus enabling various distributed, lowlatency and reliable intelligent services.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The cloud can provide powerful processing capabilities and massive storage when the edge cannot afford.

“3) powerful cloud backup, the cloud can provide powerful processing capabilities and massive storage when the edge cannot afford.”

May combine two claims.

Edge computing combined with cloud computing provides agile service response by reducing delay of data transmissions and improving response speed.

“2) agile service response, services hosted at the edge can significantly reduce the delay of data transmissions and improve the response speed;”

Probably not a substantial claim: may not pass S3 (load-bearing).

50 billion IoT devices will be connected to the Internet by 2020.

“According to a Cisco white paper [1], 50 billion IoT devices will be connected to the Internet by 2020.”

• cites [1] 2019Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based). May combine two claims.

Diversified and valuable DL services can broaden the commercial value of edge computing and accelerate its deployment and growth.

“5) diversified and valuable DL services can broaden the commercial value of edge computing and accelerate its deployment and growth.”

May combine two claims.

Nearly 850 ZB of data will be generated each year outside the cloud by 2021, while global data center traffic is only 20.6 ZB.

“Cisco estimates that nearly 850 Zettabytes (ZB) of data will be generated each year outside the cloud by 2021, while global data center traffic is only 20.6 ZB [2].”

• cites [2] 2019Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

DL is adopted in a myriad of applications and corroborates its superior performance in computer vision, natural language processing, and AI.

“With respect to CV, NLP, and AI, DL is adopted in a myriad of applications and corroborates its superior performance [78].”

• cites [78] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S4 (based).

Executing real-time video analytic tasks in the cloud often incurs high bandwidth consumption, unexpected latency, and reliability issues.

“Unfortunately, executing these tasks in the cloud often incurs high bandwidth consumption, unexpected latency, and reliability issues.”

Classified as probably not a substantial claim. May combine two claims.

Edge nodes are expected to have computing and caching capabilities and to provide high-quality network connection and computing services near end devices.

“Edge nodes are expected to have computing and caching capabilities and to provide high-quality network connection and computing services near end devices.”

Classified as probably not a substantial claim.

The edge computing architecture, because it covers a large number of distributed edge devices, can be utilized to better serve DL.

“Nonetheless, the edge computing architecture, on account of it covers a large number of distributed edge devices, can be utilized to better serve DL.”

Probably not a substantial claim: may not pass S4 (based).

A large number of GPUs, TPUs, or FPGAs are required to be deployed in the cloud to process DL service requests.

“Currently, a large number of GPUs, TPUs, or FPGAs are required to be deployed in the cloud to process DL service requests.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

Edge computing systems with advanced and excellent microservice architecture are the future development direction for DL services with complex configuration and intensive resource requirements.

“For DL services with complex configuration and intensive resource requirements, edge computing systems with advanced and excellent microservice architecture are the future development direction.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

SDN allows rapid deployment of services, network programmability and multi-tenancy support through decoupling of control and data planes, centralized programmable control planes, and standardized application programming interface.

“Software-defined Networking (SDN) allows rapid deployment of services, network programmability and multi-tenancy support, through three key innovations [72]: 1) Decoupling of control planes and data planes; 2) Centralized and programmable control planes; 3) Standardized application programming interface.”

• cites [72] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

LAVEA attaches edge nodes to the same access point or BS as the end devices, ensuring that services can be as ubiquitous as Internet access.

“For example, LAVEA [106] attaches edge nodes to the same access point or BS as well as the end devices, which ensure that services can be as ubiquitous as Internet access.”

• cites [106] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

There are two common ways to perform distributed training: data parallelism and model parallelism.

“There are two common ways to perform distributed training, i.e., data parallelism and model parallelism [95]- [98] as illustrated in Fig. 10.”

• cites [95]- [98]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Typical network optimization methodologies rely on fixed mathematical models; it is difficult to accurately model rapidly changing edge network environments and systems.

“Typical network optimization methodologies rely on fixed mathematical models; however, it is difficult to accurately model rapidly changing edge network environments and systems.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

VM is better at isolating while container provides easier deployment of repetitive tasks.

“In general, VM is better at isolating while container provides easier deployment of repetitive tasks [69].”

• cites [69] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

The focus of the collaborative end-edge-cloud paradigm is achieving the optimal balance of equipment energy consumption, server loads, transmission and execution delays.

“The focus of this collaborative paradigm is not only the successful completion of tasks but also achieving the optimal balance of equipment energy consumption, server loads, transmission and execution delays.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

A large number of end devices, edge nodes, and cloud data centers are scattered and envisioned to be connected by edge computing networks, and these distributed devices can potentially be powerful contributors once DL training jumps out of the cloud.

“Coincidentally, a large number of end devices, edge nodes, and cloud data centers, are scattered and envisioned to be connected by virtue of edge computing networks.These distributed devices can potentially be powerful contributors once the DL training jumps out of the cloud.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The edge computing architecture will become an important complement to the cloud, even replacing the role of the cloud in some scenarios.

“The edge computing architecture will become an important complement to the cloud, even replacing the role of the cloud in some scenarios.”

Probably not a substantial claim: may not pass S4 (based).

Edge computing cannot provide as many resources for DL services as the cloud does.

“Edge computing cannot provide that resources for DL services as the cloud does.”

May combine two claims.

VideoEdge implements an end-edge-cloud hierarchical architecture to help achieve load balancing concerning analytical tasks while maintaining high analysis accuracy.

“To fully exploit the bond function of the edge, VideoEdge [109] implements an end-edge-cloud hierarchical architecture to help achieve load balancing concerning analytical tasks while maintaining high analysis accuracy.”

• cites [109] sourceNot checked — no full text of the source

May combine two claims.

At the end level, video capture devices such as smartphones and surveillance cameras are responsible for video capture, media data compression, image preprocessing, and image segmentation.

“At the end level, video capture devices, such as smartphones and surveillance cameras are responsible for video capture, media data compression [102], image preprocessing, and image segmentation [103].”

• cites [102] sourceNot checked — no full text of the source
• cites [103] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Training a large DL model generally requires a lot of computation resources, even thousands of CPUs.

“Training a large DL model generally requires a lot of computation resources, even thousands of CPUs are required to train a large-scale DL model.”

Classified as probably not a substantial claim.

The combination of edge computing and DL to form high-performance Edge DL services requires coordinated integration of computing, networking and communication resources.

“The combination of edge computing and DL to form high-performance Edge DL services requires the coordinated integration of computing, networking and communication resources, as depicted in Fig. 8.”

Probably not a substantial claim: may not pass S4 (based).

Edge devices typically have limited computing power or power consumption compared to the cloud.

“Certainly, edge devices typically have limited computing power or power consumption compared to the cloud.”

May combine two claims.

Container virtualization for edge computing can effectively reduce workload execution time with high performance and storage requirements, and can deploy a large number of services in a scalable and straightforward fashion.

“Container virtualization for edge computing can effectively reduce the workload execution time with high performance and storage requirements, and can also deploy a large number of services in a scalable and straightforward fashion [70].”

• cites [70] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Edge nodes have caching ability, which can improve response time by caching popular contents.

“In addition, edge nodes also have the ability to cache, which can improve the response time by caching popular contents.”

Probably not a substantial claim: may not pass S3 (load-bearing).

A container consists of a single file that includes an application and execution environment with all dependencies, enabling efficient service handoff to cope with user mobility.

“A container consists of a single file that includes an application and execution environment with all dependencies, which makes it enable efficient service handoff to cope with user mobility [71].”

• cites [71] sourceNot checked — no full text of the source

May combine two claims.

Deploying edge nodes to perform computation tasks accelerates task processing while ensuring accuracy.

“Therefore, by deploying edge nodes to perform the computation task, the task processing can be accelerated while ensuring accuracy.”

May combine two claims.

Edge DL services typically require high bandwidth, low latency, and dynamic network configuration.

“Further, Edge DL services typically require high bandwidth, low latency, and dynamic network configuration, while Software-defined Networking (SDN) allows rapid deployment of services, network programmability and multi-tenancy support, through three key innovations [72]:”

May combine two claims.

NFV/SDN can enhance the interoperability of edge computing infrastructure; edge nodes can be efficiently orchestrated and integrated with cloud data centers.

“For example, with the support of NFV/SDN, edge nodes can be efficiently orchestrated and integrated with cloud data centers [73].”

• cites [73] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

The edge computing community has not yet reached a consensus on standardized definitions, architectures and protocols of edge computing.

“However, the edge computing community has not yet reached a consensus on the standardized definitions, architectures and protocols of edge computing [23].”

• cites [23] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S4 (based).

Real-time video analytic requires high computation and storage resources.

“In general, applying DL for it requires high computation and storage resources.”

May combine two claims.

The cloud needs to integrate different well-trained DL models to achieve global knowledge because distributed model training performance on an edge node may be significantly impaired due to its local knowledge.

“Since the distributed model training performance on an edge node may be significantly impaired due to its local knowledge, the cloud needs to integrate different well-trained DL models to achieve global knowledge.”

• cites [102] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

A hybrid hierarchical architecture for real-time video analytics is divided into three levels: end, edge, and cloud.

“In this section, as depicted in Fig. 11, we summarize related works as a hybrid hierarchical architecture, which is divided into three levels: end, edge, and cloud.”

Classified as probably not a substantial claim. May combine two claims.

DL can extract valuable information from complex network data and make adaptive decisions, achieving intelligent maintenance and management.

“DL is expected to deal with this problem: when faced with complex and cumbersome network information, DL can rely on its powerful learning and reasoning ability to extract valuable information from data and make adaptive decisions, achieving intelligent maintenance and management accordingly.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

The combination of DL and edge computing is not straightforward and requires a comprehensive understanding of DL models and edge computing features for design and deployment.

“Therefore, the combination of DL and edge computing is not straightforward and requires a comprehensive understanding of DL models and edge computing features for design and deployment.”

May combine two claims.

At the cloud level, the cloud is responsible for the integration of DL models among the edge layer and updating parameters of distributed DL models on edge nodes.

“At the cloud level, the cloud is responsible for the integration of DL models among the edge layer and updating parameters of distributed DL models on edge nodes [102].”

• cites [102] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

Distributed training can facilitate the training process by taking full advantage of parallel servers.

“Nonetheless, distributed training can facilitate the training process by taking full advantage of parallel servers.”

May combine two claims.

When the edge is unable to provide the service confidently, the cloud can use its powerful computing power and global knowledge for further processing and assist the edge nodes to update DL models.

“When the edge is unable to provide the service confidently (e.g., detecting objects with low confidence), the cloud can use its powerful computing power and global knowledge for further processing and assist the edge nodes to update DL models.”

Classified as probably not a substantial claim.

Preprocessing raw data at the edge before transmitting to cloud can reduce intermediate data size and network traffic, alleviating backbone network pressure.

“If these operations can largely reduce the size of intermediate data and hence reduce the network traffic, the pressure of backbone networks can be alleviated.”

May combine two claims.

Customized edge DL frameworks can help match edge platform and DL models, exploit underlying hardware for performance and power, and orchestrate and maintain DL services automatically.

“customized edge DL frameworks can help efficiently 1) match edge platform and DL models; 2) exploit underlying hardware in terms of performance and power; 3) orchestrate and maintain DL services automatically.”

• cites [179] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

For the protection of sensitive home data, the data processing of smart home systems must rely on edge computing.

“For the protection of sensitive home data, the data processing of smart home systems must rely on edge computing.”

• AssumptionRests on an assumption: Data processing of smart home systems must rely on edge computing for protection of sensitive home data.

May combine two claims.

LAVEA achieves speedup ranging from 1.3× to 4× against running in local and 1.2× to 1.7× against client-cloud configuration.

“Have a speedup ranging from 1.3× to 4× (1.2× to 1.7×) against running in local (clientcloud confguration)”

• cites [106] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

DeepThings improves throughput by 1.7×-2.2× and speeds up CNN inference by 1.7×-3.5×.

“improve throughput by 1.7×-2.2×and speedup CNN inference by 1.7×-3.5×”

• cites [156] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

DeepThings reduces memory footprint by more than 68% without sacrificing accuracy.

“Reduce memory footprint by more than 68% without sacrificing accuracy”

• cites [156] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

V2V communication technology can be taken advantage of to further connect vehicles, either as an edge node or an end device managed by DRL-based control policies.

“Besides, Vehicle-to-Vehicle (V2V) communication technology can be taken advantaged to further connect vehicles, either as an edge node or an end device managed by DRL-based control policies [112].”

• cites [112] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

A DRL-based approach is first utilized to solve the optimization problem for enhancing the holistic system performance in the integrated IoV framework.

“Since this system involves multi-dimensional control, a DRLbased approach is first utilized to solve the optimization problem for enhancing the holistic system performance.”

• cites [110] sourceNot checked — no full text of the source

May combine two claims.

IONN divides a DL model into multiple partitions and uploads them sequentially, allowing the edge node to incrementally build the DL model and execute offloaded partial DL computation before the entire model is uploaded.

“IONN divides a DL model, prepared for uploading, into multiple partitions, and uploads them to the edge node in sequential.The edge node, receiving the partitioned models, incrementally builds the DL model as each partitioned model arrives, while being able to execute the offloaded partial DL computation even before the entire DL model is uploaded.”

• cites [175] sourceNot checked — no full text of the source

May combine two claims.

In vertical collaboration, the edge performs data pre-processing and preliminary learning, and intermediate data are transmitted to the cloud for further DL computation.

“A natural choice of collaboration is the edge performs data pre-processing and preliminary learning, when the DL tasks are offloaded.Then, the intermediate data, viz., the output of edge architectures, are transmitted to the cloud for further DL computation [176].”

• cites [176] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S2 (checkable).

The power, energy efficiency, memory footprint limitation of IoT edge devices should also be considered.

“Besides, the power, energy efficiency, memory footprint limitation of IoT edge devices [116] should also be considered.”

• cites [116] sourceNot checked — no full text of the source

May combine two claims.

Edge computing is deployed to optimize indoor positioning systems and home intrusion monitoring so that they can get lower latency than using cloud computing as well as better accuracy.

“Like use cases in [118], [119], edge computing is deployed to optimize indoor positioning systems and home intrusion monitoring so that they can get lower latency than using cloud computing as well as the better accuracy.”

• cites [118] sourceNot checked — no full text of the source
• cites [119] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Neurosurgeon profiles all layers of a DNN on end device and edge node to generate performance prediction models, and uses them with wireless conditions and server load to partition the DNN at the best point for latency or energy.

“In [12], all layers of a DNN are profiled on the end device and the edge node in terms of the data and computation characteristics, in order to generate performance prediction models.Based on these prediction models, wireless conditions and server load levels, the proposed Neurosurgeon evaluates each candidate point in terms of end-to-end latency or mobile energy consumption and partition the DNN”

• cites [12] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

The combination of DL and edge computing can make intelligent services become more various and powerful.

“Further, the combination of DL and edge computing can make these intelligent services become more various and powerful.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

DL tasks are usually computationally intensive and require large memory footprints.

“DL tasks are usually computationally intensive and requires large memory footprints.”

May combine two claims.

In intelligent factories, edge computing is conducive to expand the computation resources, network bandwidth, and storage capacity of the cloud to the IoT edge, as well as realizing resource scheduling and data processing during manufacturing and production.

“In intelligent factories, edge computing is conducive to expand the computation resources, the network bandwidth, and the storage capacity of the cloud to the IoT edge, as well as realizing the resource scheduling and data processing during manufacturing and production [114].”

• cites [114] sourceNot checked — no full text of the source

May combine two claims.

FTP can fuse layers and partition them vertically in a grid fashion, minimizing required memory footprint of participating edge devices regardless of number of partitions and devices, while reducing communication and task migration cost.

“In contrast to only partitioning the DNN by layers as in [12], FTP can fuse layers and partitions them vertically in a grid fashion, hence minimizing the required memory footprint of participated edge devices regardless of the number of partitions and devices, while reducing communication and task migration cost as well.”

• cites [156] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The original intention of applying edge computing in smart cities is more due to cost and efficiency considerations.

“The original intention of applying edge computing in smart cities is more due to cost and efficiency considerations.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

ALOHA formulates a toolflow to automate model design, optimize model configuration, and automate model porting.

“ALOHA [182] formulates a toolflow: 1) Automate the model design.It generates the optimal model configuration by taking into account the target task, the set of constraints and the target architecture; 2) Optimize the model configuration.It partitions the DL model and accordingly generates architecture-aware mapping information between different inference tasks and the available resources.3) Autom”

• cites [182] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S4 (based).

Intermediate DNN output can be designed much smaller than raw input, drastically reducing network traffic between end and edge (or edge and cloud).

“Nevertheless, the intermediate output, e.g., high-resolution surveillance video streams, should be carefully designed much smaller than the raw input, therefore drastically reducing the network traffic required between the end and the edge (or the edge and the cloud).”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

DL models should be further customized to fit in the resource-constrained edge, while carefully treating the trade-off between the inference accuracy and the execution latency of them.

“Therefore, DL models should be further customized to fit in the resource-constrained edge, while carefully treating the trade-off between the inference accuracy and the execution latency of them.”

Classified as probably not a substantial claim. May combine two claims.

The outstanding performance of DL models is inseparable from the support of high-level hardware, and it is difficult to deploy them in the edge with limited resources.

“Certainly, the outstanding performance of DL models is inseparable from the support of high-level hardware, and it is difficult to deploy them in the edge with limited resources.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

With exponential growth of IoT edge devices, remotely managing evolving DL models and continuously evaluating these models are necessary.

“Nonetheless, with the exponential growth of IoT edge devices, 1) how to remotely manage evolving DL models and 2) how to continuously evaluate these models for them are necessary.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Vehicles can be connected to improve safety, enhance efficiency, reduce accidents, and decrease traffic congestion in transportation systems.

“It is envisioned that vehicles can be connected to improve safety, enhance efficiency, reduce accidents, and decrease traffic congestion in transportation systems [110].”

• cites [110] sourceNot checked — no full text of the source

May combine two claims.

MoDNN executes DL in a local distributed mobile computing system over WLAN, partitioning each DNN layer into slices executed layer-by-layer to increase parallelism and reduce memory footprint, significantly accelerating DL computation.

“MoDNN, proposed in [177], executes DL in a local distributed mobile computing system over a Wireless Local Area Network (WLAN).Each layer of DNNs is partitioned into slices to increase parallelism and to reduce memory footprint, and these slices are executed layerby-layer.By the execution parallelism among multiple end devices, the DL computation can be significantly accelerated.”

• cites [177] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

For important data sources, data safety and privacy protection should be addressed.

“Moreover, for important data sources, data safety and privacy protection should be addressed.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The combination of DL and edge computing endows robots the ability of dynamic visual servicing and enables efficient music cognition system.

“For instance, it endows robots the ability of dynamic visual servicing [120] and enables efficient music cognition system [121].”

• cites [120] Tian, 2018 sourceNot checked — no full text of the source
• cites [121] sourceNot checked — no full text of the source

Classified as probably not a substantial claim.

The natural characteristic of geographically distributed data sources in cities requires an edge computing-based paradigm to offer location-awareness and latency-sensitive monitoring and intelligent control.

“The natural characteristic of geographically distributed data sources in cities requires an edge computingbased paradigm to offer location-awareness and latencysensitive monitoring and intelligent control.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Edge computing provides low-latency, high-speed communication and fast-response services for vehicles, making automatic driving possible.

“On one hand, edge computing provides low-latency, highspeed communication and fast-response services for vehicles, making automatic driving possible.”

May combine two claims.

The architecture in [122] can support latency-sensitive applications on end devices and perform slightly latency-tolerant tasks efficiently on edge nodes, while large-scale DL models responsible for deep analysis are hosted on the cloud.

“This architecture can not only support latency-sensitive applications on end devices but also perform slightly latency-tolerant tasks efficiently on edge nodes, while large-scale DL models responsible for deep analysis are hosted on the cloud.”

• cites [122] sourceNot checked — no full text of the source

May combine two claims.

FedPAQ provides near-optimal theoretical guarantees for strongly convex and non-convex loss functions and empirically demonstrates communication-computation tradeoff.

“FedPAQ [204] simultaneously incorporates these features and provides near-optimal theoretical guarantees for both strongly convex and non-convex loss functions, while empirically demonstrating the communication-computation tradeoff.”

• cites [204] Reisizadeh, 2019 sourceNot checked — no full text of the source

May combine two claims. The quote only partly states this claim.

FL can deal with non-IID training data, limited communication, unbalanced contribution, and privacy/security challenges in edge computing networks.

“To summarize, FL can deal with several key challenges in edge computing networks: 1) Non-IID training data.”

• AssumptionRests on an assumption: Training data at edge devices may be distributed non-uniformly.

Probably not a substantial claim: may not pass S3 (load-bearing).

Zoo provides a concise DSL to enable easy and type-safe composition of DL services.

“Zoo [184] provides a concise Domain-specific Language (DSL) to enable easy and type-safe composition of DL services.”

• cites [184] sourceNot checked — no full text of the source

May combine two claims.

ECO uses a graph-based overlay network to model and track pipelines and dependencies and map them to geographically distributed analytic engines, distributing DL computation as needed to manage cost and performance while supporting engine heterogeneity and discontinuous operations.

“ECO [185] uses a graph-based overlay network approach to 1) model and track pipelines and dependencies and then 2) map them to geographically distributed analytic engines ranging from small edge-based engines to powerful multi-node cloud-based engines.By this means, DL computation can be distributed as needed to manage cost and performance, while also supporting other practical situations, such as”

• cites [185] Talagala, ECO: Harmonizing edge and cloud with ML/DL orchestrationNot checked — no full text of the source

May combine two claims.

[218] improves model accuracy by 4% and accelerates training by 2×.

“Training acceleration: 2×; Model accuracy improvement: 4%”

• cites [218] Xu, 2019 sourceNot checked — no full text of the source

May combine two claims.

Edge devices may suffer from higher latency, lower transmission rate and intermittent connections, hindering gradients exchanging between DL models of different edge devices.

“Besides, in practical, edge devices may suffer from higher latency, lower transmission rate and intermittent connections, and therefore further hindering the gradients exchanging between DL models belong to different edge devices.”

• AssumptionRests on an assumption: Edge devices may suffer from higher latency, lower transmission rate and intermittent connections.

Probably not a substantial claim: may not pass S4 (based).

Edge architecture with many edge nodes with modest computing resources can alleviate network pressure by processing data or training at themselves.

“Naturally, the edge architecture, which consists of a large number of edge nodes with modest computing resources, can cater for alleviating the pressure of networks by processing the data or training at themselves.”

May combine two claims.

No single DL library or hardware platform can entirely outperform others, and loading the DL model may take more time than executing it.

“Experimental results show that no single DL library or hardware platform can entirely outperform others, and loading the DL model may take more time than that of executing it.”

• cites [186] Zhang, pCAMP: Performance comparison of machine learning packages on the edges sourceNot checked — no full text of the source
• cites [54] Ignatov, 2018 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

[203] reduces communication cost by two orders of magnitude.

“Communication cost reduction: by two orders of magnitude”

• cites [203] Konečný, 2016 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S4 (based).

Merging all data into the cloud might violate privacy issues.

“In addition, merging all data into the cloud might violate privacy issues.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

FL is emerging and is promised to address issues of DL training at edge.

“Nonetheless, FL is emerging and is promised to address these issues.”

May combine two claims.

Decentralized SGD method in [189] is designed for seismic imaging and cannot be generalized for future DL training because communication cost for large-scale DL models is extremely high.

“However, this proposed method is designed for seismic imaging application and can not be generalized for future DL training, since the communication cost for training large scale DL models is extremely high.”

• cites [189] sourceNot checked — no full text of the source

May combine two claims.

Continually transmitting massive data from distributed end devices or edge nodes to the cloud consumes significant communication resources and aggravates wireless and backbone networks.

“Besides, a significant amount of communication resources will be consumed, and hence aggravating wireless and backbone networks if massive data are required to be continually transmitted from distributed end devices or edge nodes to the cloud.”

May combine two claims.

Federated learning must be robust to the unexpected drop out of edge devices, or else once a device loses its connection, the synchronization of FL in one round will be failed.

“FL must be robust to the unexpectedly drop out of edge devices, or else once a device loses its connection, the synchronization of FL in one round will be failed.”

Probably not a substantial claim: may not pass S3 (load-bearing).

[223] leverages blockchain to exchange and verify model updates of local training, affecting learning completion latency.

“Leverage blockchain to exchange and verify model updates of local training Learning completion latency”

• cites [223] Kim, 2018 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Combining Blockchain and FL as BlockFL is proposed in [223] to realize locally global model updating at each edge device rather than a specific server, ensuring device malfunction cannot affect other local updates when updating the global model, and an appropriate reward mechanism for stimulating edge devices to participate in FL.

“Therefore, in [223], combining Blockchain and FL as BlockFL is proposed to realize 1) locally global model updating at each edge device rather a specific server, ensuring device malfunction cannot affect other local updates when updating the global model; 2) appropriate reward mechanism for stimulating edge devices to participate in FL.”

• cites [223] Kim, 2018 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

Laplace noises are added to possibly exposed training data to enhance training data privacy assurance in [196].

“Hence, Laplace noises [197] are added to these possibly exposed training data for enhancing the training data privacy assurance.”

• cites [196] sourceNot checked — no full text of the source
• cites [197] sourceNot checked — no full text of the source

May combine two claims.

The server should not trust edge devices completely, since devices with abnormal behavior can forge or poison their training data, which results in worthless model updates, and hence harming the global model.

“However, the server also should not trust edge devices completely, since devices with abnormal behavior can forge or poison their training data, which results in worthless model updates, and hence harming the global model.”

Probably not a substantial claim: may not pass S3 (load-bearing).

[211] accelerates convergence by 62.4%.

“Convergence acceleration: 62.4%”

• cites [211] Hu, 2019 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

The total wall-clock time of FL includes computation time and communication time of all clients.

“Specifically, the total wall-clock time of FL includes not only the computation time but also the communication time of all clients.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

OpenEI defines each DL algorithm as a four-element tuple to evaluate Edge DL capability of target hardware and can select a matched model for a specific edge platform online.

“OpenEI [183] defines each DL algorithm as a four-element tuple ¡Accuracy, Latency, Energy, Memory Footprint¿ to evaluate the Edge DL capability of the target hardware platform.Based on such tuple, OpenEI can select a matched model for a specific edge platform based on different Edge DL capabilities in an online manner.”

• cites [183] Zhang, 2019 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

A standard testbed for Edge DL is missing, hindering study of edge architectures for DL.

“Nonetheless, a standard testbed for Edge DL is missing, which hinders the study of edge architectures for DL.”

May combine two claims.

Large-scale distributed training at edge avoids transmitting bulky raw dataset to cloud but inevitably introduces communication cost for gradients exchanging between edge devices.

“Though largescale distributed training at edge evades transmitting bulky raw dataset to the cloud, the communication cost for gradients exchanging between edge devices is inevitably introduced.”

Probably not a substantial claim: may not pass S3 (load-bearing).

[201] uses trimmed mean as robust aggregation to improve Top 1 accuracy against data poisoning.

“Use the trimmed mean as a robust aggregation Top 1 accuracy against data poisoning”

• cites [201] Xie, 2019 sourceNot checked — no full text of the source

May combine two claims.

Compared to DQL, the experiment results indicate that Double-DQL can save more energy and achieve higher training efficiency.

“Compared to DQL, the experiment results indicate that Double-DQL can save more energy and achieve higher training efficiency.”

• cites [257] sourceNot checked — no full text of the source

May combine two claims.

Though offloading computation tasks to edge nodes can enhance the processing efficiency of the computation tasks, the reliability of offloading suffers from the potentially low quality of wireless environments.

“Though offloading computation tasks to edge nodes can enhance the processing efficiency of the computation tasks, the reliability of offloading suffers from the potentially low quality of wireless environments.”

Probably not a substantial claim: may not pass S3 (load-bearing).

There are many potential areas where DL can be applied, including online content streaming, routing and traffic control, etc.

“In this regard, there are many potential areas where DL can be applied, including online content streaming [285], routing and traffic control [286] [287], etc.”

• cites [285] sourceNot checked — no full text of the source
• cites [286] sourceNot checked — no full text of the source
• cites [287] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S4 (based).

The key issue of extending DL from the cloud to the edge of the network is how to devise and develop edge computing architecture to achieve the best performance of DL training and inference under the multiple constraints of networking, communication, computing power, and energy consumption.

“In summary, the key issue of extending DL from the cloud to the edge of the network is: under the multiple constraints of networking, communication, computing power, and energy consumption, how to devise and develop edge computing architecture to achieve the best performance of DL training and inference.”

May combine two claims.

The performance of DNN-based methods is bounded by fixed optimization algorithms and is not self-adapted.

“Therefore, the performance of DNN-based methods bounds by fixed optimization algorithms and is not self-adapted.”

May combine two claims.

DRL-based security solutions can provide secure offloading from the edge device to the edge node against jamming attacks or protect user location privacy and the usage pattern privacy.

“DRL-based security solutions can provide secure offloading (from the edge device to the edge node) to against jamming attacks [261] or protect user location privacy and the usage pattern privacy [270].”

• cites [261] sourceNot checked — no full text of the source
• cites [270] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

DL solutions do not rely entirely on accurate modeling of networks and devices, so finding a scenario where DL can be applied is not the most important concern.

“However, since DL solutions do not rely entirely on accurate modeling of networks and devices, finding a scenario where DL can be applied is not the most important concern.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Among all these works related to learning-based optimization methods, DL-based approaches have advantages over others when multiple edge nodes and radio channels are available for computation offloading.

“Among all these works related to learning-based optimization methods, DL-based approaches have advantages over others when multiple edge nodes and radio channels are available for computation offloading.”

Probably not a substantial claim: may not pass S2 (checkable).

Taking the limited computing capabilities of edge devices into consideration, partitioning a large-scale DL model and allocating these segments to different edge devices for training may be a more feasible and practical solution.

“Hence, taking the limited computing capabilities of edge devices (at least for now) into consideration, partitioning a large-scale DL model and allocating these segments to different edge devices for training may be a more feasible and practical solution.”

• AssumptionRests on an assumption: DL training at the edge is currently mainly limited by the weak performance of edge devices and the fact that most Edge DL frameworks or libraries still do not support training.

Probably not a substantial claim: may not pass S4 (based).

If applying DL to optimize realtime edge computing networks, the training and inference of DL models or DRL algorithms may bring certain side effects, such as additional bandwidth consumed by training data transmission and latency of DL inference.

“Besides, if applying DL to optimize realtime edge computing networks, the training and inference of DL models or DRL algorithms may bring certain side effects, such as the additional bandwidth consumed by training data transmission and the latency of DL inference.”

Probably not a substantial claim: may not pass S3 (load-bearing).