Dynamic Adaptive DNN Surgery for Inference Acceleration on the Edge
The DOI matches its Crossref record; 22 of the 23 references it cites were found in a registry or an index.
Abstract
Recent advances in deep neural networks (DNNs) have substantially improved the accuracy and speed of a variety of intelligent applications. Nevertheless, one obstacle is that DNN inference imposes heavy computation burden to end devices, but offloading inference tasks to the cloud causes transmission of a large volume of data. Motivated by the fact that the data size of some intermediate DNN layers is significantly smaller than that of raw input data, we design the DNN surgery, which allows partitioned DNN processed at both the edge and cloud while limiting the data transmission. The challenge is twofold: (1) Network dynamics substantially influence the performance of DNN partition, and (2) State-of-the-art DNNs are characterized by a directed acyclic graph (DAG) rather than a chain so that partition is greatly complicated. In order to solve the issues, we design a Dynamic Adaptive DNN Surgery (DADS) scheme, which optimally partitions the DNN under different network condition. Under the lightly loaded condition, DNN Surgery Light (DSL) is developed, which minimizes the overall delay to process one frame. The minimization problem is equivalent to a min-cut problem so that a globally optimal solution is derived. In the heavily loaded condition, DNN Surgery Heavy (DSH) is developed, with the objective to maximize throughput. However, the problem is NP-hard so that DSH resorts an approximation method to achieve an approximation ratio of 3. Real-world prototype based on self-driving car video dataset is implemented, showing that compared with executing entire the DNN on the edge and cloud, DADS can improve latency up to 6.45 and 8.08 times respectively, and improve throughput up to 8.31 and 14.01 times respectively.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474
Going deeper with convolutions
Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke, V., & Rabinovich, A. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298594 · Cited by 48,562
YOLO9000: Better, Faster, Stronger
Redmon, J., & Farhadi, A. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.690 · Cited by 19,892 · cited 4×
“For example, the input data size of tiny YOLOv2 [4] is 0 .”
BackgroundInfluentialInception-v4, Inception-ResNet and the Impact of Residual Connections on Learning
Szegedy, C., Ioffe, S., Vanhoucke, V., & Alemi, A. (2017) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v31i1.11231 · Cited by 4,636 · cited 1×
“6 shows the DAG of the pure inception v4 network [10] in Fig.”
BackgroundAn experimental comparison of min-cut/max- flow algorithms for energy minimization in vision
Boykov, Y., & Kolmogorov, V. (2004) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2004.60 · Cited by 4,577
MAUI: making smartphones last longer with code offload
Cuervo, E., Balasubramanian, A., Cho, D.-K., Wolman, A., Saroiu, S., Chandra, R., & Bahl, P. (2010) · Proceedings of the 8th international conference on Mobile systems, applications, and services · doi:10.1145/1814433.1814441 · Cited by 2,281
Dynamic adaptive streaming over HTTP -: standards and design principles
Stockhammer, T. (2011) · Proceedings of the second annual ACM conference on Multimedia systems · doi:10.1145/1943552.1943572 · Cited by 1,221 · cited 1×
“This can be realized by a method similar to HTTP DASH [11].”
MethodDeep neural networks for small footprint text-dependent speaker verification
Variani, E., Lei, X., McDermott, E., Moreno, I. L., & Gonzalez-Dominguez, J. (2014) · 2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp.2014.6854363 · Cited by 1,035
Distributed Deep Neural Networks Over the Cloud, the Edge and End Devices
Teerapittayanon, S., McDanel, B., & Kung, H. T. (2017) · 2017 IEEE 37th International Conference on Distributed Computing Systems (ICDCS) · doi:10.1109/icdcs.2017.226 · Cited by 843
(2011) · Cited by 713
Cited by
Items 101–200 of 493
Carnero, A., Martín, C., Torres, D. R., Garrido, D., Díáz, M., & Rubio, B. (2021) · IEEE Access · doi:10.1109/access.2021.3110291 · Cited by 29
Liquid Software-Based Edge Intelligence for Future 6G Networks
Yang, T., Qin, M., Cheng, N., Xu, W., & Zhao, L. Y. (2022) · IEEE Network · doi:10.1109/mnet.011.2000654 · Cited by 29
Distributed Inference in Resource-Constrained IoT for Real-Time Video Surveillance
Khan, M. A., Hamila, R., Erbad, A., & Gabbouj, M. (2022) · IEEE Systems Journal · doi:10.1109/jsyst.2022.3198711 · Cited by 29
“For instance, Jankowski et al. [3] improved the model accuracy, Hu et al. [5] reduced inference latency but ignores energy consumption, and Zhao et al. [7] reduced the memory footprint.”
BackgroundMethodInfluentialAdaptive Federated Learning for Non-Convex Optimization Problems in Edge Computing Environment
Qiao, D., Liu, G., Guo, S., & He, J. (2022) · IEEE Transactions on Network Science and Engineering · doi:10.1109/tnse.2022.3185116 · Cited by 28
“The application of the combination of AI and EC mainly focuses on model training [24] and model inference [25].”
Horizontally Distributed Inference of Deep Neural Networks for AI-Enabled IoT
Rodríguez, I., Campos, C., & Fdez‐Riverola, F. (2023) · Sensors · doi:10.3390/s23041911 · Cited by 28
“Seminal studies along these lines, e.g., Neurosurgeon [66], IONN [67], DADS [68], and QDMP [69], while still far from being sufficient approaches to achieving the objective stated above, have been instrumental in laying the foundations of DNN task partitioning and offloading, identifying key concepts and resulting in te…”
Shi, L., Xu, Z., Sun, Y., Shi, Y., Fan, Y., & Ding, X. (2021) · Peer-to-Peer Networking and Applications · doi:10.1007/s12083-021-01223-1 · Cited by 28
“For DNNs those are characterized by a directed acyclic graph (DAG) rather than a chain, Hu et al. in [20] proposed a partition method based on graph min-cut method, and proposed the partition algorithms under the light workload and the heavy workload respectively.”
MethodEdgeCI: Distributed Workload Assignment and Model Partitioning for CNN Inference on Edge Clusters
Chen, Y., Luo, T., Fang, W. W., & Xiong, N. N. (2024) · ACM Transactions on Internet Technology · doi:10.1145/3656041 · Cited by 27
“To alleviate discrepancy between the intensive resource requirements of CNNs and the limited computing powers of IoT devices, many researchers have explored a lot of optimization technologies, including model compression [12, 13], dynamic inference [14, 15], and edge-cloud synergy [16–18].”
BackgroundKilcioglu, E., Mirghasemi, H., Stupia, I., & Vandendorpe, L. (2021) · IEEE Access · doi:10.1109/access.2021.3084689 · Cited by 27
EdgeLD: Locally Distributed Deep Learning Inference on Edge Device Clusters
Feng, X., Fang, W. W., Xu, W., Wang, Q., Ma, X., & Ding, Y. (2020) · IEEE International Conference on High Performance Computing and Communications; IEEE International Conference on Smart City; IEEE International Conference on Data Science and Systems (HPCC/SmartCity/DSS) · doi:10.1109/hpcc-smartcity-dss50907.2020.00078 · Cited by 27
“Nevertheless, the inference performance is highly dependent on unpredictable network conditions [5].”
BackgroundMobility-Included DNN Partition Offloading from Mobile Devices to Edge Clouds
Tian, X., Zhu, J., Xu, T., & Li, Y. (2021) · Sensors · doi:10.3390/s21010229 · Cited by 26
“[27] proposed a system partitioned DNN to be processed at both the edge cloud and the central cloud while limiting the data transmission, and it can optimally partition the DNN under the lightly loaded condition and heavily loaded condition, respectively.”
MethodHu, Y., Xu, X., Duan, L., Bilal, M., Wang, Q., & Dou, W. (2024) · IEEE Transactions on Fuzzy Systems · doi:10.1109/tfuzz.2024.3412971 · Cited by 25
Computation Offloading Scheduling for Deep Neural Network Inference in Mobile Computing
Duan, Y., & Wu, J. (2021) · International Workshop on Quality of Service · doi:10.1109/iwqos52092.2021.9521304 · Cited by 24
“Prior research has discussed the optimal partition strategy for cooperative DNN inference [5], [6].”
BackgroundMethodYang, L., Shen, X., Zhong, C., & Liao, Y. (2022) · Journal of Parallel and Distributed Computing · doi:10.1016/j.jpdc.2022.09.005 · Cited by 24
Pacheco, R. G., Couto, R. S., & Simeone, O. (2023) · Journal of Network and Computer Applications · doi:10.1016/j.jnca.2023.103679 · Cited by 24
“Reference Classifies images Employs early-exit DNNs Calibrates the model Analyzes the impact of calibration Considers a realistic adaptive offloading scenario [11] Yes No No No No [12] Yes Yes Yes No Yes [15] Yes Yes Yes No Yes [18, 19] No Yes No No No [20, 21] No Yes No No No [25] No No No No Yes [26] Yes Yes No No N…”
BackgroundMethodSmartCooper: Vehicular Collaborative Perception with Adaptive Fusion and Judger Mechanism
Zhang, Y., An, H., Fang, Z., Xu, G., Zhou, Y. H., Chen, X., & Fang, Y. M. (2024) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra57147.2024.10610199 · Cited by 24
“Regarding image data, contemporary cameras, exemplified by Google’s autonomous vehicles (AVs), are capable of capturing a staggering 750 megabytes of data per second [20].”
MethodDeep Learning Video Analytics on Edge Computing Devices
Tan, T., & Cao, G. (2021) · IEEE International Conference on Sensing, Communication, and Networking (SECON) · doi:10.1109/secon52354.2021.9491614 · Cited by 24
“Although researchers have proposed offloading techniques [7]–[9] which can determine jobs to be offloaded to the edge server or processed locally to minimize the completion time, none of them considers the memory constraint and its effect on scheduling.”
BackgroundATPP : A Mobile App Prediction System Based on Deep Marked Temporal Point Processes
Yang, K., Zhao, X., Zou, J., & Du, W. (2023) · ACM Transactions on Sensor Networks · doi:10.1145/3582555 · Cited by 23
“There are some systems that are focused on the DNN inference on the edge devices [24, 25, 45, 53, 61].”
BackgroundDNN Partitioning for Inference Throughput Acceleration at the Edge
Feltin, T., Marchó, L., Cordero-Fuertes, J.-A., Brockners, F., & Clausen, T. (2023) · IEEE Access · doi:10.1109/access.2023.3244497 · Cited by 23
“Methods optimizing inference throughput include DNN surgery [12], which uses the lower size of intermediary DNN layer outputs to partition the inference computation between the edge and the cloud.”
MethodThe Case for Hierarchical Deep Learning Inference at the Network Edge
Al-Atat, G., Fresa, A., Behera, A. P., Moothedath, V. N., Gross, J., & Champati, J. P. (2023) · doi:10.1145/3597062.3597278 · Cited by 23
PerDNN: Offloading Deep Neural Network Computations to Pervasive Edge Servers
Jeong, H.-J., Lee, H.-J., Shin, K. Y., Yoo, Y. H., & Moon, S. (2020) · IEEE International Conference on Distributed Computing Systems (ICDCS) · doi:10.1109/icdcs47774.2020.00114 · Cited by 24
“Recently, Chuang Hu et al. proposed a partitioning algorithm applicable to DAG-formed DNNs based on the min-cut algorithm [22].”
MethodSequential Offloading for Distributed DNN Computation in Multiuser MEC Systems
Wang, F., Cai, S., & Lau, V. K. N. (2023) · IEEE Internet of Things Journal · doi:10.1109/jiot.2023.3279271 · Cited by 22
A cloud-edge collaborative task scheduling method based on model segmentation
Zhang, C., Chen, J., Li, W., Sun, H., Geng, Y., Zhang, T., Ji, M., & Fu, T. (2024) · Journal of Cloud Computing Advances Systems and Applications · doi:10.1186/s13677-024-00635-7 · Cited by 22
“Finally, this paper compares the time complexity of CECMS algorithm with an adaptive DNN inference acceleration framework with end–edge–cloud collaborative computing algorithm [40] (ADC) and a Dynamic Adaptive DNN Surgery method (DADS) [21].”
MethodKrouka, M., Elgabli, A., Issaid, C. B., & Bennis, M. (2021) · IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC) · doi:10.1109/pimrc50174.2021.9569707 · Cited by 22
Dynamic DNN Decomposition for Lossless Synergistic Inference
Zhang, B., Tian, X., Zhang, H., Li, T., Zhu, S., & Gu, J. (2021) · IEEE International Conference on Distributed Computing Systems Workshops (ICDCSW) · doi:10.1109/icdcsw53096.2021.00010 · Cited by 22
JMSNAS: Joint Model Split and Neural Architecture Search for Learning over Mobile Edge Networks
Tian, Y., Zhang, Z., Yang, Z., & Yang, Q. (2022) · IEEE International Conference on Communications workshops · doi:10.1109/iccworkshops53468.2022.9814499 · Cited by 22
Artificial intelligence in edge devices
Singh, A., Saini, K., Nagar, V., Aseri, V., Sankhla, M. S., Pandit, P. P., & Chopade, R. L. (2022) · Advances in computers · doi:10.1016/bs.adcom.2022.02.013 · Cited by 22
Manasi, S. D., Snigdha, F. S., & Sapatnekar, S. S. (2020) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2020.2995135 · Cited by 22
“In [8] and [9], partitioning is used to optimize overall delay or throughput for delay critical applications of deep neural network (DNN) (e.”
MethodZhang, P., Huang, F., Wu, D., Yang, B., Yang, Z., & Tan, L. (2023) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2023.3240884 · Cited by 21
“The delay model of the modules containing subgraphs in ResNet101 can be abstracted as a DAG: G = ( V, E ) , in which the optimal partitioning point is found such that the total delay is minimized under unpredictable network dynamics [14].”
BackgroundMethodInfluentialNear-optimal and learning-driven task offloading in a 5G multi-cell mobile edge cloud
Xia, Q., Lou, Z., Xu, W., & Xu, Z. (2020) · Computer Networks · doi:10.1016/j.comnet.2020.107276 · Cited by 21
Learning the Optimal Partition for Collaborative DNN Training With Privacy Requirements
Zhang, L., & Xu, J. (2021) · IEEE Internet of Things Journal · doi:10.1109/jiot.2021.3127715 · Cited by 20
Optimization of Offloading Policies for Accuracy-Delay Tradeoffs in Hierarchical Inference
Beytur, H. B., Aydin, A. G., Veciana, G. A. D., & Vikalo, H. (2024) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom52122.2024.10621325 · Cited by 20
“Unlike DNN Partitioning [9], [10], Hierarchical Inference does not (partially) offload every task to the server, which helps improve the system’s responsiveness.”
BackgroundSynergy: Towards On-Body AI via Tiny AI Accelerator Collaboration on Wearables
Gong, T., Jang, S. Y., Acer, U. G., Kawsar, F., & Min, C. (2025) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2025.3564314 · Cited by 20
“12(c); for example, the model inference in the second run does not need to wait for the completion of the pipeline in the first run when the AI accelerator is idle and the sensor data for the second run is ready.”
InfluentialNeuroMessenger: Towards Error Tolerant Distributed Machine Learning Over Edge Networks
Wang, S., & Zhang, X. (2022) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom48880.2022.9796695 · Cited by 20
Pipelining Split Learning in Multi-Hop Edge Networks
Wei, W., Lin, Z., Li, T., Li, X., & Chen, X. (2026) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2026.3684802 · Cited by 19
“Moreover, Hu et al. [24] optimally partition DNNs between edge and cloud computing based on dynamic network conditions, which addresses the computing-communication trade-off.”
Narukulla, N. (2024) · Journal of Electrical Systems · Cited by 19
Towards scalable and efficient Deep-RL in edge computing: A game-based partition approach
Dai, H., Wu, J., Wang, Y., & Xu, C. (2022) · Journal of Parallel and Distributed Computing · doi:10.1016/j.jpdc.2022.06.006 · Cited by 19
“However, compared to existing partition methods which either adopt heuristic methods [24,28,33] or graph-based methods [20], Coknight bears some distinct features, which are short of adequate considerations in the existing studies: 1) Coknight is fairly efficient by accomplishing the partition in quadratic time; 2) Co…”
BackgroundMethodInfluentialKreß, F., Sidorenko, V., Schmidt, P., Hoefer, J., Hotfilter, T., Walter, I., Harbaum, T., & Becker, J. (2023) · Computer Networks · doi:10.1016/j.comnet.2023.109759 · Cited by 19
A Review of Application Layer Communication Protocols for the IoT Edge Cloud Continuum
Kampars, J., Tropins, D., & Matisons, R. (2021) · International Scientific Conference on Information Technology and Management Science of Riga Technical University (ITMS) · doi:10.1109/itms52826.2021.9615332 · Cited by 19
PICO: Pipeline Inference Framework for Versatile CNNs on Diverse Mobile Devices
Yang, X., Xu, Z., Qi, Q., Wang, J., Sun, H., Liao, J., & Guo, S. (2023) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2023.3265111 · Cited by 19
“DADS [27] proposed a novel algorithm to partition DNN with graph structure using a min-cut algorithm.”
MethodGetting the Best Out of Both Worlds: Algorithms for Hierarchical Inference at the Edge
Moothedath, V. N., Champati, J. P., & Gross, J. (2024) · IEEE Transactions on Machine Learning in Communications and Networking · doi:10.1109/tmlcn.2024.3366501 · Cited by 19
“Following this idea, significant research work has been done that includes DNN partitioning for more general DNN structures under different network settings [11], [32] and using heterogeneous EDs [12], among others.”
BackgroundEdge-Cloud Collaborated Object Detection via Difficult-Case Discriminator
Cao, Z., Li, Z., Chen, Y., Pan, H., Hu, Y., & Liu, J. (2023) · IEEE International Conference on Distributed Computing Systems (ICDCS) · doi:10.1109/icdcs57875.2023.00062 · Cited by 18
“1-top left) [6], [23], [24].”
BackgroundOptimizing DNN training with pipeline model parallelism for enhanced performance in embedded systems
Maruf, M. A., Azim, A., Auluck, N., & Sahi, M. (2024) · Journal of Parallel and Distributed Computing · doi:10.1016/j.jpdc.2024.104890 · Cited by 18
Hu, S., Lu, Z., Xu, X., Deng, R., Du, X., & Duan, Q. (2025) · Journal of Industrial Information Integration · doi:10.1016/j.jii.2025.100955 · Cited by 18
“DADS Hu, Bao, Wang and Liu (2019) formulates different model partition optimization problems for lightly and heavily loaded conditions.”
Edge-Assisted Real-Time Instance Segmentation for Resource-Limited IoT Devices
Xie, Y., Guo, Y., Mi, Z., Yang, Y., & Obaidat, M. S. (2022) · IEEE Internet of Things Journal · doi:10.1109/jiot.2022.3199921 · Cited by 18
“Hu et al. [36] dynamically partitioned DNNs with respect to different network conditions.”
BackgroundAdaEE: Adaptive Early-Exit DNN Inference Through Multi-Armed Bandits
Pacheco, R. G., Shifrin, M., Couto, R. S., Menasché, D. S., Hanawal, M. K., & Campista, M. E. M. (2023) · International Conference on Communications · doi:10.1109/icc45041.2023.10279243 · Cited by 18
Cloud–edge cooperation for meteorological radar big data: a review of data quality control
Hu, Z., Xu, X., Zhang, Y., Tang, H., Cheng, Y., Qian, C., & Khosravi, M. R. (2021) · Complex & Intelligent Systems · doi:10.1007/s40747-021-00581-w · Cited by 18
“[43] designed a dynamic adaptive ground DNN partitioning scheme with optimized partitioning ofDNNs under different network conditions and significantly improved throughput at the edge and in the cloud (Table 3).”
MethodEnergy and Time-Aware Inference Offloading for DNN-based Applications in LEO Satellites
Chen, Y., Zhang, Q., Zhang, Y., Ma, X., & Zhou, A. (2023) · IEEE International Conference on Network Protocols (ICNP) · doi:10.1109/icnp59255.2023.10355644 · Cited by 17
“Furthermore, numerous excellent experiments have already performed the partitioning of DNNs into subtasks based on layers [13], [14].”
BackgroundWang, H., Li, T., Zhang, M., Li, Q., Cui, H., Jiang, Y., & Yuan, Z. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3465434 · Cited by 17
“The feature map is subsequently transmitted to the cloud and further processed using the subsequent part of the DNN. DADS [70] further discussed the method of splitting more complex model with a a DAG (Directed Acyclic Graph) topology.”
DNN Inference Acceleration Based on Adaptive Task Partitioning and Offloading in Embedded VEC
Li, C., Liu, S., Jiang, K., Yang, M., Zhang, Z., Wang, B., Zhao, L., Chen, C., & Wan, S. (2025) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3725734 · Cited by 17
“The speed is controlled at [20,100] km/h.”
InfluentialFEVA: A Federated Video Analytics Architecture for Networked Smart Cameras
Hu, C., Lu, R., & Wang, D. (2021) · IEEE Network · doi:10.1109/mnet.001.2100261 · Cited by 17
“Edge-cloud computing architectures [8] were proposed where the edge devices will conduct initial computing of a few DNN/CNN layers to reduce the amount of data transmission, see Fig.”
MethodSelf-citationResiliNet: Failure-Resilient Inference in Distributed Neural NetworksPreprint
Yousefpour, A., Nguyen, B., Devic, S., Wang, G., Kreidieh, A., Löbel, H., Bayen, A. M., & Jue, J. P. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2002.07386 · Cited by 18
“…neural networks is less explored, primarily due to the emerging application scenarios that need ongoing and long inference tasks (Teerapittayanon et al., 2017; Morshed et al., 2017; Liu et al., 2018; Tao and Li, 2018; Hu et al., 2019; Dey et al., 2019). b. Neural Network Fault Tolerance .”
BackgroundCarbonCP: Carbon-Aware DNN Partitioning with Conformal Prediction for Sustainable Edge IntelligencePreprint
Ke, H., Jin, W., & Wang, H. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2404.16970 · Cited by 18
“While previous work has leveraged DNN partitioning primarily for improving latency and inference accuracy [5, 6, 7], to the best of our knowledge, it has not yet been exploited for carbon reduction within the context of edge computing.”
MethodSplitwise: Collaborative Edge–Cloud Inference for LLMs via Lyapunov-Assisted DRL
Younesi, A., Maryan, A. S., Oustad, E., Samani, Z. N., Ansari, M., & Fahringer, T. (2025) · IEEE/ACM International Conference on Utility and Cloud Computing (UCC) · doi:10.1145/3773274.3774267 · Cited by 16
Service Provisioning in Edge-Cloud Continuum: Emerging Applications for Mobile Devices
Rodrigues, D. O., Souza, A. M. D., Braun, T. I., Maia, G., Loureiro, A. A. F., & Villas, L. A. (2023) · Journal of Internet Services and Applications · doi:10.5753/jisa.2023.2913 · Cited by 16
“Another type of EC approach to enhance mobile gaming experience is UAV-assisted EC (Kim et al., 2018b; Hu et al., 2019b).”
BackgroundMethodInfluentialKim, H. C., Choi, J. S., Kim, J., & Ko, J. H. (2024) · Future Generation Computer Systems · doi:10.1016/j.future.2024.01.006 · Cited by 16
Edge-Cloud Collaborated Object Detection via Bandwidth Adaptive Difficult-Case Discriminator
Cao, Z., Cheng, Y., Zhou, Z., Chen, Y., Hu, Y., Lu, A., Liu, J., & Li, Z. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3474743 · Cited by 16
Collaborative Inference in Resource-Constrained Edge Networks: Challenges and Opportunities
Ng, N., Souza, A., Diggavi, S. N., Suri, N., Abdelzaher, T., Towsley, D., & Shenoy, P. J. (2024) · IEEE Military Communications Conference (MILCOM) · doi:10.1109/milcom61039.2024.10773876 · Cited by 16
“One critical consideration when using this approach is determining the amount of processing to be performed by the sensing node [8], [9].”
The Case for Adaptive Deep Neural Networks in Edge Computing
McNamee, F., Dustdar, S., Kilpatrick, P., Shi, W., Spence, I. T. A., & Varghese, B. (2021) · IEEE ... International Conference on Cloud Computing · doi:10.1109/cloud53861.2021.00017 · Cited by 16
“The impact of network conditions for DNN repartitioning has been considered [30].”
BackgroundDo we need sophisticated system design for edge-assisted augmented reality?
Meng, J., Kong, Z. J., Hu, Y., Choi, M. G., & Lal, D. (2022) · International Workshop on Edge Systems, Analytics and Networking (EdgeSys) · doi:10.1145/3517206.3526267 · Cited by 16
“…the offloading latency which was extensively studied in recent years is to partition the DNN model for inference between the mobile client and the server and find the optimal partitioning point that minimizes the end-to-end latency, e.g., that has a small model feature size ( e.g., [8, 10, 12–15]).”
BackgroundReal-time DNN Model Partitioning for QoE Enhancement in Mobile Vision Applications
Lim, J.-A., & Kim, Y. (2022) · IEEE International Conference on Sensing, Communication, and Networking (SECON) · doi:10.1109/secon55815.2022.9918600 · Cited by 16
“…selects a partition point to minimize per-frame E2E latency with a fixed N MHz GPU clock frequency, (iii) DADS ( N ) : an algorithm proposed in [12] which selects a partition point considering processed fps and E2E latency depending on whether the system is under-loaded or overloaded with a…”
BackgroundInfluentialMind Your Heart: Stealthy Backdoor Attack on Dynamic Deep Neural Network in Edge Computing
Dong, T., Zhang, Z., Qiu, H., Zhang, T., Li, H., & Wang, T. (2023) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom53939.2023.10229092 · Cited by 16
“One of the state-of-the-art approaches to solve the above issue is to transform a vanilla DNN into a dynamic DNN to deploy on edge devices [6]–[8].”
BackgroundOfpCNN: On-Demand Fine-Grained Partitioning for CNN Inference Acceleration in Heterogeneous Devices
Yang, L., Zheng, C., Shen, X., & Xie, G. (2023) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2023.3321755 · Cited by 16
“On the basis of [23], the collaborative reasoning method of model compression in [24] reduces the complexity of the algorithm and improves the execution efficiency of the model partition algorithm on resource-constrained terminal devices.”
MethodSocialized Learning: A Survey of the Paradigm Shift for Edge Intelligence in Networked Systems
Wang, X., Zhao, Y., Qiu, C., Hu, Q., & Leung, V. C. M. (2024) · IEEE Communications Surveys & Tutorials · doi:10.1109/comst.2024.3482978 · Cited by 17
“This challenge has been addressed by Hu et al. [187] through the development of an adaptive scheme.”
Enable Pipeline Processing of DNN Co-inference Tasks In the Mobile-Edge Cloud
Hu, S., Dong, C., & Wen, W. (2021) · International Conference on Computer and Communication Systems (ICCCS) · doi:10.1109/icccs52626.2021.9449178 · Cited by 16
“We also notice that recent work formulates CNN models as directed acyclic graph(e.g. GoogleNet [11], ResNet [12]).”
Tian, X., Xu, P., Shen, Y., & Shao, Y. (2023) · Peer-to-Peer Networking and Applications · doi:10.1007/s12083-023-01564-z · Cited by 15
“Literature [8] proposes to regard DNN as a DAG structure and proposes two approximate algorithms to solve the problem of minimizing latency for networks with small and large network loads, respectively.”
BackgroundChen, X., Lin, C., & Lin, B. (2023) · IEEE Transactions on Network Science and Engineering · doi:10.1109/tnse.2023.3321089 · Cited by 15
Li, Z., Yu, H., Fan, G., Zhang, J., & Xu, J. (2024) · Journal of Parallel and Distributed Computing · doi:10.1016/j.jpdc.2024.104850 · Cited by 15
A Benchmark for ML Inference Latency on Mobile Devices
Li, Z., Paolieri, M., & Golubchik, L. (2024) · International Workshop on Edge Systems, Analytics and Networking (EdgeSys) · doi:10.1145/3642968.3654818 · Cited by 15
“…networks on resource-constrained mobile devices is an essential task across a broad range of applications, e.g.,: (1) collaborative inference [9, 11], where a deep neural network is partitioned between a mobile device and a cloud server for cooperative processing – accurate latency…”
BackgroundJanus: Collaborative Vision Transformer Under Dynamic Network Environment
Jiang, L., Fu, S. D., Zhu, Y., & Li, B. (2025) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom55648.2025.11044626 · Cited by 15
Accelerating End-Cloud Collaborative Inference via Near Bubble-Free Pipeline Optimization
Gao, L., Liu, J., Xu, H., Xu, S., Ma, Q., & Liusheng, H. (2025) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom55648.2025.11044632 · Cited by 15
“In practice, end devices often need to process continuous inference tasks, necessitating efficient scheduling strategies to enhance the throughput of inference systems [2].”
InfluentialCombination of Task Allocation and Approximate Computing for Fog-Architecture-Based IoT
Yu, W., Najafi, A., Huang, Y., & García-Ortiz, A. (2020) · IEEE Internet of Things Journal · doi:10.1109/jiot.2020.3040892 · Cited by 15
Joint compressing and partitioning of CNNs for fast edge-cloud collaborative intelligence for IoT
Zhang, W., Wang, N., Li, L., & Wei, T. (2022) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2022.102461 · Cited by 15
GradMDM: Adversarial Attack on Dynamic Networks
Pan, J., Foo, L. G., Zheng, Q., Fan, Z., Rahmani, H., Ke, Q., & Liu, J. (2023) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2023.3263619 · Cited by 15
“This can be dangerous in some safety-critical applications such as self-driving vehicles, where a timely response is important [59].”
BackgroundDeep Learning on Mobile Devices With Neural Processing Units
Tan, T., & Cao, G. (2023) · Computer · doi:10.1109/mc.2022.3215780 · Cited by 15
“Although there are some existing works 10, 16 on this topic, they rely on uncalibrated confidence scores, which is not effective for some DNNs, and a fixed confidence score threshold is used to determine which frames should be offloaded.”
BackgroundEfficient Pipeline Collaborative DNN Inference in Resource-Constrained UAV Swarm
Ren, W., Qu, Y., Qin, Z., Dong, C., Zhou, F., Zhang, L., & Wu, Q. (2024) · IEEE Wireless Communications and Networking Conference · doi:10.1109/wcnc57260.2024.10570535 · Cited by 14
“These studies can be divided into three main categories: cloud computing based UAV collaborative inference [5]–[7], edge server assisted UAV collaborative inference [8]–[10], and collaborative DNN inference in UAV swarm [11], [12].”
BackgroundReducing inference energy consumption using dual complementary CNNs
Kinnas, M., Violos, J., Kompatsiaris, I. Y., & Papadopoulos, S. (2024) · Future Generation Computer Systems · doi:10.1016/j.future.2024.107606 · Cited by 14
SLICE: Energy-Efficient Satellite-Ground Co-Inference via Layer-Wise Scheduling Optimization
Chen, Y., Zhang, Q., Xing, R., Li, Y., Ma, X., Zhang, Y., Zhou, A., & Wang, S. (2025) · IEEE Transactions on Services Computing · doi:10.1109/tsc.2025.3577451 · Cited by 14
“For instance, Hu et al. [21] introduced DNN surgery, enabling simultaneous segmentation inference at the edge and cloud to reduce latency.”
Li, J., Liang, W., Li, Y., Xu, Z., & Jia, X. (2021) · Conference on Local Computer Networks · doi:10.1109/lcn52139.2021.9524928 · Cited by 14
“The essential differences of our work in this paper from the aforementioned study [8] lie in the following aspects.”
BackgroundMethodResultInfluentialEdge–IoT Computing and Networking Resource Allocation for Decomposable Deep Learning Inference
Yang, Y.-T. C., & Wei, H. (2022) · IEEE Internet of Things Journal · doi:10.1109/jiot.2022.3222461 · Cited by 14
“Research[16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28] shows that sometimes decomposing a DL application task into several parts and then running one part of the inference on the IoT device and the other parts on the more powerful computing devices can effectively…”
BackgroundSecurity-Communication-Computation Tradeoff of Split Decisions for Edge Intelligence
Lyu, X., Liu, J., Ren, C., & Nan, G. (2023) · IEEE Wireless Communications · doi:10.1109/mwc.014.2200438 · Cited by 14
Memory-aware and context-aware multi-DNN inference on the edge
Cox, B., Birke, R., & Chen, L. Y. (2022) · Pervasive and Mobile Computing · doi:10.1016/j.pmcj.2022.101594 · Cited by 14
Measuring and Controlling Split Layer Privacy Leakage Using Fisher InformationPreprint
Maeng, K., Guo, C., Kariyappa, S., & Suh, G. E. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2209.10119 · Cited by 14
CoActo: CoActive Neural Network Inference Offloading with Fine-grained and Concurrent Execution
Bin, K., Park, J.-K., Park, C., Kim, S., & Lee, K. (2024) · International Conference on Mobile Systems, Applications, and Services · doi:10.1145/3643832.3661885 · Cited by 13
“Motivated by this, many researchers propose expanding the collaborative inference approach to multi-path models [10], adopting early-exit [16], uploading model [13].”
BackgroundZhai, X., Fu, S., Yi, C., Liu, Z., Dong, C., & Tan, C. W. (2025) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2025.3629117 · Cited by 13
“Collaborative inference [14], [15] o ff ers a potential solution to reduce latency and energy consumption by partitioning the DNN into front-end and back-end components, as shown in Fig.”
AI Strategy for Stake Cloud Computing and Edge Computing: A State of the art survey
Aishwarya, R., & Mathivanan, G. (2021) · International Conference on Electronics, Communication and Aerospace Technology (ICECA) · doi:10.1109/iceca52323.2021.9676013 · Cited by 13
Dynamic Path Based DNN Synergistic Inference Acceleration in Edge Computing Environment
Meng, Z., Zhou, B., Wang, H., Dong, F., & Zhao, W. (2021) · IEEE International Conference on Parallel and Distributed Systems (ICPADS) · doi:10.1109/icpads53394.2021.00076 · Cited by 13
Server load and network-aware adaptive deep learning inference offloading for edge platforms
Ahn, J., Lee, Y., Ahn, J., & Ko, J. (2022) · Internet of Things · doi:10.1016/j.iot.2022.100644 · Cited by 13
DECC: Delay-Aware Edge-Cloud Collaboration for Accelerating DNN Inference
Zhuang, Z., Chen, J., Xu, W., Qi, Q., Guo, S., Wang, J., Lu, L., Yang, H., & Liao, J. (2024) · IEEE Transactions on Emerging Topics in Computing · doi:10.1109/tetc.2024.3404551 · Cited by 13
“Other partitioningalgorithms,suchas[18],[19],searchforasuitable partitioning strategy when the three stages are serialized.”
InfluentialFailure-Resilient Distributed Inference With Model Compression Over Heterogeneous Edge Devices
Wang, L., Li, L., Xu, L., Peng, X., & Fei, A. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3419831 · Cited by 13
“…been devoted to mitigating the computational bottleneck at single edge devices by coordinating multiple devices to jointly perform intelligent tasks via proper model/data partition and workload distribution, facilitated by controllable inter-device communication [3], [4], [5], [6], [7], [9], [18].”
BackgroundGeng, H., Zeng, D., Li, Y., Gu, L., Chen, Q., & Li, P. (2024) · IEEE Internet of Things Journal · doi:10.1109/jiot.2024.3443289 · Cited by 13
“2) Dynamic Adaptive DNN Surgery (DADS) [34]: DADS splits the DNN model into two parts and executes them at end device and edge server, respectively.”
Enabling Edge Artificial Intelligence via Goal-oriented Deep Neural Network Splitting
Binucci, F., Merluzzi, M., Banelli, P., Strinati, E. C., & Lorenzo, P. D. (2024) · International Symposium on Wireless Communication Systems · doi:10.1109/iswcs61526.2024.10639178 · Cited by 13
“The authors of [12] and [13] consider DNN partitioning schemes to minimize the latency and the user equipment energy consumption, respectively.”
BackgroundeDeepSave: Saving DNN Inference using Early Exit During Handovers in Mobile Edge Environment
Ju, W., Yuan, D., Bao, W., Ge, L., & Zhou, B. B. (2021) · ACM Transactions on Sensor Networks · doi:10.1145/3447267 · Cited by 13
Self-citationAccelerate Cooperative Deep Inference via Layer-wise Processing Schedule Optimization
Wang, N., Duan, Y., & Wu, J. (2021) · International Conference on Computer Communications and Networks (ICCCN) · doi:10.1109/icccn52240.2021.9522274 · Cited by 13
“Hu et al. further considered the optimal partition by considering the fact the many SOTA DNNs have a DAG architecture and proposed a min-cut formulation to solve latency minimization using a max-flow approach [8].”
BackgroundMethodInfluentialTowards Edge-Cloud Collaborative Machine Learning: A Quality-aware Task Partition Framework
Zheng, Z., Li, Y., Han, S., Wang, L., & Xia, F. (2022) · ACM International Conference on Information and Knowledge Management (CIKM) · doi:10.1145/3511808.3557080 · Cited by 13
“Task Partition for machine-learning service has been intensively researched in both cloud and edge computing community, including two categories based on partitioning content: function-based [6, 7, 42, 43] and data-based [44–47].”
BackgroundDynaSpa: Exploiting Spatial Sparsity for Efficient Dynamic DNN Inference on Devices
Liu, R., Leng, Y., Tian, S., Hu, S., Chen, R., & Yao, S. (2024) · ACM Conference on Embedded Networked Sensor Systems (SenSys) · doi:10.1145/3666025.3699348 · Cited by 12
DNN Partitioning for Cooperative Inference in Edge Intelligence: Modeling, Solutions, Toolchains
Hao, Y., Ding, N., Xia, W., Ge, H., & Xu, L. (2026) · ACM Computing Surveys · doi:10.1145/3786145 · Cited by 14
“Recurrent neural networks ( RNNs ) and generative adversarial networks ( GANs ), these architectures have received comparatively less attention.”
Onboard Optimization and Learning: A Survey
Pavel, M. I., Hu, S., Pratama, M., & Kowalczyk, R. (2026) · IEEE Access · doi:10.1109/access.2026.3664956 · Cited by 12
“…with reported speedups exceeding one order of magnitude and energy reductions approaching an order of magnitude on heterogeneous edge platforms, whereas degraded bandwidth consistently shifts optimal execution toward earlier partition points or fully local inference [133], [146], [148].”
InfluentialAn Adaptive Task Migration Scheduling Approach for Edge‐Cloud Collaborative Inference
Zhang, B., Li, Y., Zhang, S., Zhang, Y., & Zhu, B. (2022) · Wireless Communications and Mobile Computing · doi:10.1155/2022/8804530 · Cited by 12
Reliable DNN Partitioning for UAV Swarm
Zhao, M., Zhang, X., Meng, Z., & Hou, X. (2022) · International Wireless Communications and Mobile Computing Conference (IWCMC) · doi:10.1109/iwcmc55113.2022.9824945 · Cited by 12
Yang, Y.-T. C., & Wei, H. (2022) · IEEE Transactions on Network and Service Management · doi:10.1109/tnsm.2022.3220521 · Cited by 12
Metadata
- Authors
- Hu, C., Bao, W., Wang, D., & Liu, F.
- Affiliations
- Hong Kong Polytechnic University (HK)
- The University of Sydney (AU)
- Venue
- IEEE INFOCOM 2019 - IEEE Conference on Computer Communications
- Publisher
- IEEE
- Published
- 2019-04
- Pages
- 1423-1431
- OpenAlex
- W2920031528
- Cited by
- 416
- References
- 23
- Citations by year
- 2026: 29 · 2025: 61 · 2024: 65 · 2023: 92 · 2022: 62 · 2021: 72 · 2020: 26 · 2019: 9
- FWCI
- 36.52
- Type
- conference paper
- Language
- en
- Open access
- closed
- Licence
- Locations
- IEEE Conference on Computer Communications (INFOCOM) (publishedVersion)
- Rare & Special e-Zone (The Hong Kong University of Science and Technology) (submittedVersion)
- Topics
- Physical Sciences › Computer Science › Computer Networks and Communications › IoT and Edge/Fog Computing
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Hardware and Architecture › Cloud Computing and Remote Desktop Technologies
- Keywords
- DNN partitioning, cloud-edge collaboration, inference acceleration, dynamic network conditions, directed acyclic graphs, minimum cut, approximation algorithms, throughput maximization, latency reduction, data transmission reduction
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.