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 301–400 of 493
Multi-User Mobile Augmented Reality with ID-Aware Visual Interaction
Cai, X., Yang, Z., Dong, L., Ma, Q., Miao, X., & Liu, Z. (2023) · ACM Transactions on Sensor Networks · doi:10.1145/3623638 · Cited by 4
“DADS[18] allows adaptive partitioned DNN processed at both the edge and cloud under network dynamics while limiting the data transmission.”
BackgroundResource and Bandwidth-Aware Video Analytics with Adaptive Offloading
Zhang, L., Zhong, Y., Liu, J., & Cui, L. (2023) · IEEE International Conference on Mobile Ad-Hoc and Sensor Systems (MASS) · doi:10.1109/mass58611.2023.00021 · Cited by 4
“Moreover, model partitioning is often adopted to adjust the amount of the offloaded computation by choosing an appropriate partitioning point, from which the model is divided into two parts and executed at separate sides [6], [9], [10].”
MethodMDP: Model Decomposition and Parallelization of Vision Transformer for Distributed Edge Inference
Wang, W., Zhang, Y., Jin, Y., Han, T., & Chen, L. (2023) · International Conference on Mobility, Sensing and Networking (MSN) · doi:10.1109/msn60784.2023.00086 · Cited by 4
Lyu, Y., Liu, L., Wang, X., Fan, Z., Wang, J., & Gao, G. (2025) · Machine Learning and Knowledge Extraction · doi:10.3390/make7040117 · Cited by 4
“EdgeFlow [20] designed a new distributed inference mechanism for general Directed Acyclic Graph (DAG) structured DNN models.”
Optimal DNN Inference Delay Minimization for Chain-Structured Roadside Edge Networks
Wan, X., Ji, T., Guan, X., Zhu, A., & Ye, F. (2023) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2023.3290019 · Cited by 4
“For example, [3] analyzed the characteristics of DNN layers to determine the partition point, while [4], [5] introduced graph-based optimization.”
MethodCollaborative non-chain DNN inference with multi-device based on layer parallel
Zhang, Q., Sun, S., Luo, J., Liu, M., Li, Z., Yang, H., & Wang, Y. (2023) · Digital Communications and Networks · doi:10.1016/j.dcan.2023.11.004 · Cited by 4
A Survey on Securing Image-Centric Edge Intelligence
Li, T., Hu, H., Gabbouj, M., Ye, Q., Xiang, Y., Li, J., & Li, L. (2025) · ACM Transactions on Multimedia Computing Communications and Applications · doi:10.1145/3700792 · Cited by 4
“Some studies [39, 61] adopt “bottleneck” approach, where a squeeze stage reduces dimensions or channels to make the model more compact, and an expansion stage restores them to maintain or improve accuracy.”
Legion: Tailoring Grouped Neural Execution Considering Heterogeneity on Multiple Edge Devices
Choi, K., Lee, S., Kang, B. W., & Park, Y. (2021) · IEEE International Conference on Computer Design: VLSI in Computers and Processors (ICCD) · doi:10.1109/iccd53106.2021.00067 · Cited by 4
Performance Efficient Layer-aware DNN Inference Task Scheduling in GPU Cluster
Geng, H., Zeng, D., & Li, Y. (2022) · Globecom · doi:10.1109/globecom48099.2022.10001581 · Cited by 4
“As mentioned in recent study [15], for a DNN inference task, the DNN model can be described as DAG G =< L,E >, where the vertexes set L indicates the DNN layers of the DNN model.”
BackgroundMethodTian, X., Xu, P., Hui, G., & Meng, H. (2023) · Wireless Communications and Mobile Computing · doi:10.1155/2023/3835297 · Cited by 4
A Dynamic and Collaborative Deep Inference Framework for Human Motion Analysis in Telemedicine
Boldo, M., Damiano Carra, B., Quaglia, D., & Bombieri, N. (2023) · IEEE International Conference on Edge Computing and Communications (EDGE) · doi:10.1109/edge60047.2023.00043 · Cited by 4
A Model Stealing Attack Against Multi-Exit Networks
Li, P., Lv, P., Chen, K., Zhang, S., Cai, Y., & Fan, X. (2025) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49660.2025.10888512 · Cited by 4
“Due to the advantages of fast inference [10, 11, 12] and energy-efficient computation [13, 14, 15], stealing multi-exit networks is very appealing to attackers.”
Failout: Achieving Failure-Resilient Inference in Distributed Neural Networks.Preprint
Yousefpour, A., Nguyen, B., Devic, S., Wang, G., Kreidieh, A., Löbel, H., Bayen, A. M., & Jue, J. P. (2020) · arXiv (Cornell University) · Cited by 4
Adaptive DNN Surgery for Selfish Inference Acceleration with On-demand Edge ResourcePreprint
Yang, X., Chen, D., Qi, Q., Wang, J., Sun, H., Liao, J., & Guo, S. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2306.12185 · Cited by 4
“As raw data passes through neural layers, features are extracted and become significantly smaller in size than the original data [6], [24].”
BackgroundMethodInfluentialParipati, L. K., & Hajari, V. R. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5052124 · Cited by 4
ECSAM: An Edge-cloud Video Analytics Architecture for the Segment Anything Model
Lu, R., Shi, S., & Wang, D. (2025) · doi:10.1145/3769696.3771216 · Cited by 3
Self-citationYang, C., Li, J., Xie, E., Liu, Y., Yang, L., & Tang, D. (2025) · International Conference on Communications · doi:10.1109/icc52391.2025.11161292 · Cited by 3
“LI et al. [10] explored DNN partitioning and early exit in edge systems, while HU et al. [11] proposed a DAG-based partitioning scheme.”
Wang, T., He, X., Chen, Y., & Wang, Z. (2025) · Future Generation Computer Systems · doi:10.1016/j.future.2025.108190 · Cited by 3
Wang, G., Fan, W., Li, X., Zhou, G., Xin, L., Yu, Z., & Liu, Y. (2026) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2026.3673729 · Cited by 3
“,thetransmissionlatency,andtheedgecomputation latency: The average latency can be expressed as: Note: In our simulations, the size of the final result is negligible compared to the compressed IA, so its transmission latency can be omitted [27], [28].”
Xiao, Y., Du, Q., Cheng, W., Karagiannidis, G. K., Nallanathan, A., Guizani, M., & Wang, J. (2026) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2026.3689106 · Cited by 3
“The authors of [23] formulated a model partitioning problem for deep neural networks (DNN) as a min-cut problem, and proposed a dynamic adaptive DNN surgery scheme to accelerate the inference.”
CADAEC: Content-Aware Deployment of AI Workloads in Edge-Cloud Ecosystem
Saha, R. K., Mittal, S., Singhal, R., & Nambiar, M. (2025) · ACM/SPEC International Conference on Performance Engineering (ICPE) · doi:10.1145/3676151.3719355 · Cited by 3
“Hu et al. [9] proposed a dynamic adaptive DNN surgery scheme, optimizing DNN inference by splitting processing between the edge and cloud and adapting to varying network conditions to minimize latency and maximize throughput.”
Rising From Pieces: Effective Inference at the Edge via Robust Split ML
Weng, Y., Zheng, T., Chen, Z., Hu, M., & Luo, J. (2025) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2025.3605382 · Cited by 3
“Thisinno-vativearchitecturehasalreadydemonstrateditsversatilityacrossnumerousreal-worldapplications,fromvirtualassistants[27] andautonomousdrivingsystems[31],[32]toobjectdetection algorithms[33],[34],[35]andimmersiveaugmentedreality [36],[37].”
Demo: CollabTrans: Device-cloud Collaborative Inference Framework for Transformer-based Models
Chen, J., Xu, H., & Li, M. (2025) · International Conference on Mobile Systems, Applications, and Services · doi:10.1145/3711875.3734371 · Cited by 3
“…about privacy and long data transmission latency, while on-device deployment struggles to meet the latency demands of complex DNNs. Collaborative inference [1, 3] addresses this by splitting the model into two parts, head model and tail model, deployed separately on the device and the server.”
Accelerate Multi-view Inference with End-edge Collaborative Computing
Cheng, W., Zhang, M., Dong, F., & Fu, S. (2023) · International Conference on Computer Supported Cooperative Work in Design (CSCWD) · doi:10.1109/cscwd57460.2023.10152842 · Cited by 3
“However, existing research [20]–[22] aims to speed up the single-view inference and only requires one model partition layer to be chosen.”
BackgroundCollaborative Inference for MEC Services Based on Multimodal Deep Neural Network
Liu, Y., Wang, C., Xiao, Y., Lv, Z., Xiao, L., & Ji, X. (2023) · IEEE/CIC International Conference on Communications in China (ICCC) · doi:10.1109/iccc57788.2023.10233276 · Cited by 3
“While DAGS represents complex DNN models like GoogleNet as directed acyclic graphs and partitions the model with node granularities, it still experiences inference performance degradation due to the heterogeneity among branches of multimodal DNN model [14].”
MethodAdaptive Scheduling for Edge-Assisted DNN Serving
He, J., Yang, C., He, Z., Baig, G., & Qiu, L. (2023) · IEEE International Conference on Mobile Ad-Hoc and Sensor Systems (MASS) · doi:10.1109/mass58611.2023.00071 · Cited by 3
“If the request rate is below 110 , the on-time ratios for Batch and our algorithm are 83% and 96% , when serving ResNet50 and GoogleNet.”
BackgroundInfluentialXiang, T., Meng, Q., Zhang, J., Zhang, B., Song, W., Xie, A., & Gu, J. J. (2023) · IEEE International Conference on Robotics and Biomimetics (ROBIO) · doi:10.1109/robio58561.2023.10354654 · Cited by 3
Ji, C., Hou, P., Yu, J., Wu, Y., & Tai, Y. (2022) · International Conference on Computing, Networks and Internet of Things (CNIOT) · doi:10.1109/cniot55862.2022.00021 · Cited by 3
Online object detection task offloading in UAV ad hoc networks
Wang, S., Du, Z., Du, Y., & Chen, J. (2022) · IEEE International Conference on Unmanned Systems (ICUS) · doi:10.1109/icus55513.2022.9987114 · Cited by 3
Dash, P., Kong, Z. J., Hu, Y., Turner, C., Wolfensparger, D., Choi, M. G., Kshitij, A., & McLandrich, V. E. (2023) · International Workshop on Edge Systems, Analytics and Networking (EdgeSys) · doi:10.1145/3578354.3592870 · Cited by 3
“Existing approaches include employing image or video encoding on the offloaded frames [8, 19, 22], splitting the DNN between the client and the server and offloading the intermediate representations [11, 16, 17, 27], dividing a frame into segments to pipeline the frame transmission with DNN inference [19, 25], and ide…”
BackgroundMethodEdgeC3: Online Management for Edge-Cloud Collaborative Continuous Learning
Lin, S., Zhang, X., Li, Y., Joe‐Wong, C., Duan, J., & Chen, X. (2023) · IEEE International Conference on Sensing, Communication, and Networking (SECON) · doi:10.1109/secon58729.2023.10287414 · Cited by 3
“…a distributed computing hierarchies, where the edge and terminal devices use the shallow part of the neural network for fast and local reasoning; [14] designed a scheme to optimally partition the DNN under different network conditions; and [9] consider sharing parameters of different layers.”
BackgroundAdaptive Deep Inference Framework for Cloud-Edge Collaboration
Kou, Z., Liu, Z., Yang, X., & He, X. (2023) · International Conference on Computers, Information Processing and Advanced Education (CIPAE) · doi:10.1109/cipae60493.2023.00014 · Cited by 3
“Collaborative deep inference [10-12] splits DNNs between end device and cloud.”
BackgroundMethodLi, D., Ke, Z., & Zhou, X. (2021) · ACM Symposium on QoS and Security for Wireless and Mobile Networks (Q2SWinet) · doi:10.1145/3479242.3487322 · Cited by 3
“It is shown in [4, 6, 20, 21] that by offloading the whole or even part of the CNN model to a powerful edge server, it is feasible to satisfy the accuracy and real-time requirements of object detection.”
BackgroundEnergy-Efficient Collaborative Inference in MEC: A Multi-Agent Reinforcement Learning Based Approach
Xiao, Y., Wan, K., Liang, X., & Yang, H. (2022) · International Conference on Big Data Computing and Communications (BigCom) · doi:10.1109/bigcom57025.2022.00057 · Cited by 3
“, L}, that correspond to the computation sequence and the input-output relationship of two connected layers [19], as shown in Fig.”
MethodA Deep Reinforcement Learning Approach for DNN Collaborative Inference in Mobile Edge Computing
Li, M., Li, Y., & Huang, J. (2023) · National Conference on Electrical and Electronics Engineering (ELECO) · doi:10.1109/iceee59925.2023.00052 · Cited by 3
“Zhao et al. [9] designs a dynamic adaptive DNN surgery (DADS) scheme, and DNNs characterized by a DAG topology is optimally partitioned under different network conditions.”
BackgroundMethodInfluentialFault-Tolerant Collaborative Inference through the Edge-PRUNE FrameworkPreprint
Boutellier, J. J., Tan, B., & Nurmi, J. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2206.08152 · Cited by 3
“A graph-based modeling approach is also adopted by DADS (Hu et al., 2019), the industrial effort Auto-Split (BanitalebiDehkordi et al.”
DynaSplit: A Hardware-Software Co-Design Framework for Energy-Aware Inference on EdgePreprint
May, D., Tundo, A., Ilager, S., & Brandić, I. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2410.23881 · Cited by 3
Niu, X., Cao, X., Chen, Y., & Jin, H. (2024) · ACM Transactions on Sensor Networks · doi:10.1145/3652516 · Cited by 2
Optimal Distribution of ML Models Over Edge for Applications with High Input Frequency
Le, T. T., Taherkordi, A., Eliassen, F., & Guan, P. (2024) · IEEE International Conference on Cloud Computing Technology and Science · doi:10.1109/cloudcom62794.2024.00033 · Cited by 2
Reducing the End-to-End Latency of DNN-Based Recommendation Systems in GPU Pools
Luan, G., Pang, P., Chen, Q., Chen, C., Xu, G., Zhang, C., Zi, Y., Yu, Y., Yang, G., Zhang, L., et al. (2025) · IEEE International Parallel and Distributed Processing Symposium (IPDPS) · doi:10.1109/ipdps64566.2025.00070 · Cited by 2
“For general DNN models, some studies focus on adaptively slicing the model for efficient inference [26], [27].”
Edge-PRUNE: A Dataflow-Based Framework for Distributed Signal Processing and Machine Learning
Boutellier, J. J., Tan, B., Nurmi, J., & Bhattacharyya, S. S. (2025) · IEEE Transactions on Signal Processing · doi:10.1109/tsp.2025.3598453 · Cited by 2
“DADS [27] proposed the DAG based edge-cloud DNN inference model (ECDI) that can be used to optimize the edge-cloud partitioning of more complex than chain-like DNN structures.”
Pao-Ding: Accelerating Cross-Edge Video Analytics via Automated CNN Model Partitioning
Liang, G., Han, B., Li, R., Han, X., & Sun, Z. (2025) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2025.3618296 · Cited by 2
“DADS [5] and CNN Surgery [6] apply graph min-cut techniques to partition DAG models under dynamic network environments between edge devices and cloud servers.”
InfluentialOnline Scheduling With Trajectory Prediction for Collaborative DNN Inference in Vehicular Networks
Han, Z., Zhou, R., Tan, H. H., & Lui, J. C. S. (2025) · IEEE Transactions on Networking · doi:10.1109/ton.2025.3629759 · Cited by 2
“For communication, the wireless bandwidth capacity of each RSU is [5 , 12 .”
InfluentialLatency-aware DNN Inference Acceleration with Dynamic Model Partitioning in Vehicular Edge Computing
Li, C., Yang, M., Zhang, Z., Deng, X., & Wan, S. (2025) · ACM Transactions on Architecture and Code Optimization · doi:10.1145/3786345 · Cited by 2
Satellite–Terrestrial Collaborative Inference for IoRT: Optimizing Latency and Energy Efficiency
Liu, R., Han, S., Zhang, W., Xu, X., Sun, M., & Zhang, P. (2026) · IEEE Internet of Things Journal · doi:10.1109/jiot.2026.3661195 · Cited by 2
“DNN models primarily exist in two forms: the directed acyclic graph [30] and the chain [31].”
Li, H., Madhukumar, H., Yan, S., Wu, Y., & Simeonidou, D. E. (2026) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2026.3747853 · Cited by 2
MiCU: End-to-End Smart Home Command Understanding with Large Language Model
Han, H., Hu, K., Cai, W., Zhang, D., Qin, B., Wang, Y., Jiang, J., Xiao, Y., & Du, B. (2026) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3770855.3818446 · Cited by 2
“Finally, we enforce data minimization and governance by limiting data usage to what is necessary for model development and quality assessment, and restricting access to authorized personnel under standard security controls.”
Wang, Z., Zhao, Y., Wang, Y., Li, Y., Liu, Z., Qiu, C., Wang, X., & Hu, Q. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-25158-0_17 · Cited by 2
Cotter, J., Castiñeiras, I., O’Shea, D., & Cionca, V. (2023) · IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC) · doi:10.1109/pimrc56721.2023.10293778 · Cited by 2
“Hu et al. [11] present a ”single split” partitioning solution for DAG-based DNNs.”
BackgroundTian, G., Wang, X., Li, X., & Qin, X. (2024) · International Conference on Computing, Networks and Internet of Things (CNIOT) · doi:10.1145/3670105.3670168 · Cited by 2
“DADS [7] investigates the dynamic partition policy of the directed acyclic graph(DAG) models based on the min-cut mechanism in different network conditions, and proposed corresponding algorithm for lightly loaded condition and heavily loaded conditions.”
Collaborative Inference for Deep Neural Networks in Edge Environments
Liu, M., Gu, Y., Dong, S., Wei, L., Liu, K., Yan, Y., Song, Y., Cheng, H., Tang, L., & Zhang, S. (2024) · KSII Transactions on Internet and Information Systems · doi:10.3837/tiis.2024.07.003 · Cited by 2
“3) Based on DADS [7], a widely used scheme for DNN partition, algorithm MCP is proposed to deal with the partition for different DNN models, no matter what the topology it is.”
InfluentialEfficient Latency Optimization in Batch Processing Server Systems for Multi-User Co-Inference
Ma, Y., Gao, K., & Zhang, C. (2024) · IEEE/CIC International Conference on Communications in China (ICCC) · doi:10.1109/icccworkshops62562.2024.10693813 · Cited by 2
“Studies by [9] and [10] optimize the latency and throughput of co-inference with a first-in-first-out (FIFO) strategy.”
MethodFast Edge Resource Scaling With Distributed DNN
Γιαννακάς, Θ., Tsilimantos, D., Destounis, A., & Spyropoulos, T. (2025) · IEEE Transactions on Network and Service Management · doi:10.1109/tnsm.2025.3532365 · Cited by 2
“1), e.g., by dynamically adjusting the sent features depending on network conditions [38], [39], [40].”
QUDOS: quorum-based cloud-edge distributed DNNs for security enhanced industry 4.0
Wallis, K., Reich, C., Varghese, B., & Schindelhauer, C. (2021) · IEEE/ACM International Conference on Utility and Cloud Computing (UCC) · doi:10.1145/3468737.3494094 · Cited by 2
“They are: (i) Profiling and estimation based [11, 26], (ii) linear programming based [7, 13], (iii) DNN structural modification based [17, 30], (iv) Approximation based [9], and (v) Benchmarking based [15, 27].”
MethodHeterogeneous platform-aware workload feature recognition for edge intelligence
Li, D., Li, Y., Hu, H., Zhang, T., & Jiang, C. (2022) · Physical Communication · doi:10.1016/j.phycom.2022.101620 · Cited by 2
Resource-Efficient DNN Training and Inference for Heterogeneous Edge Intelligence in 6G
Cui, E., Zhang, W., Yang, D., Wu, W. R., & Lyu, F. (2021) · 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-dss-smartcity-dependsys53884.2021.00135 · Cited by 2
“Hu et al. [9] considered a more complicated scenario, in which DNNs are modeled as a directed acyclic graph (DAG).”
BackgroundMethodRelation Extraction from Videos Based on IoT Intelligent Collaboration Framework
Lv, J., Shen, Q., Lv, M., & Shi, L. (2022) · Mathematics · doi:10.3390/math10183308 · Cited by 2
Automated Search for Deep Neural Network Inference Partitioning on Embedded FPGA
Kreß, F., Hoefer, J., Hotfilter, T., Walter, I., Annabi, E. M. E., Harbaum, T., & Becker, J. (2023) · Communications in computer and information science · doi:10.1007/978-3-031-23618-1_37 · Cited by 2
Deep learning inference time guarantee in near future edge computing
Genda, K. (2023) · 2012 International Conference on Computing, Networking and Communications (ICNC) · doi:10.1109/icnc57223.2023.10074015 · Cited by 2
“DADS continues to monitor the network condition and determine if the inference is operated in the lightly loaded or heavily loaded condition.”
MethodDiTing: Edge Assisted Real-time ID-aware Visual Interaction for Multi-user Augmented Reality
Cai, X., Yang, Z., Liang, D., Ma, Q., Miao, X., & Liu, Z. (2023) · IEEE International Conference on Parallel and Distributed Systems (ICPADS) · doi:10.1109/icpads56603.2022.00101 · Cited by 2
“DADS [37] allows adaptive partitioned DNN processed at both the edge and cloud under network dynamics while limiting the data transmission.”
BackgroundT2C: A Multi-User System for Deploying DNNs in a Thing-to-Cloud Continuum
Hsieh, C.-Y., Venkateswaran, P., Venkatasubramanian, N., & Hsu, C.-H. (2022) · International Conference on Mobility, Sensing and Networking (MSN) · doi:10.1109/msn57253.2022.00052 · Cited by 2
“[7] considered DNNs in graph representations and proposed two partition algorithms for light and heavy workload.”
Learning to Prevent Input Leakages in the Mobile Cloud Inference
Xiang, L., Zhang, S., & Zhang, Q. (2023) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2023.3340338 · Cited by 2
“One mainstream approach is to extract features at the mobile device and to further process the features with DNNs at the backend cloud [1], [2], [3], [4], [5].”
BackgroundOffloading DNN Tasks Based on Graph Reinforcement Learning in Client-Edge-Cloud Environment
Li, M., Huang, J., & Wang, Z. (2024) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn60899.2024.10650388 · Cited by 2
“Due to the fact that the output data volume of certain intermediate layers of the DNN is significantly smaller than its original input data volume, this method has been widely used in DNN inference [6].”
MethodPoster: Computation Offloading for Precision Agriculture using Cooperative Inference
Bovee, N., Sanchez, P. R., Ho, S.-S., Bitla, S., Patapanchala, G. K., & Piccolo, S. (2024) · IEEE International Conference on Fog and Edge Computing (ICFEC) · doi:10.1109/icfec61590.2024.00023 · Cited by 2
“However, their solution can only be applied in line-architecture DNNs. Hu et al. [8] extended the cooperative AI inference to DNNs with DAG architectures.”
BackgroundBalanced segmentation of CNNs for multi-TPU inference
Villarrubia, J., Costero, L., Igual, F. D., & Olcoz, K. (2024) · The Journal of Supercomputing · doi:10.1007/s11227-024-06605-9 · Cited by 2
“Few papers address multi-segment partitioning in scenarios similar to ours.”
Adaptive Feature Compression and Resource Scheduling for End-Edge Co-Inference
Bai, T., Huang, B., Xu, Z., Hou, B., Zhao, H., & Wang, Z. (2025) · IEEE Internet of Things Journal · doi:10.1109/jiot.2025.3582220 · Cited by 2
“1) Co-Inference Without Compression: Early research efforts [21], [22], [23] mainly focused on network partitioning strategies scheduling without data compression.”
Joint Optimization of Offloading, Batching and DVFS for Multiuser Co-Inference
Xu, Y., Zhou, S., & Niu, Z. (2025) · International Conference on Communications · doi:10.1109/icc52391.2025.11161748 · Cited by 2
“However, existing studies on edge-device co-inference [3]–[5] cannot be directly applied to these DNN hardware accelerators.”
Algorithmic Enablers for Compact Neural Network Topology Hardware Design: Review and Trends
Guicquéro, W., & Verdant, A. (2020) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas45731.2020.9181005 · Cited by 2
Conflict-Resilient Incremental Offloading of Deep Neural Networks to the Edge of Smart Environment
Chen, Z., Xu, Z., Wan, J., Tian, J., Liu, L., & Zhang, Y. (2021) · Mathematical Problems in Engineering · doi:10.1155/2021/9985006 · Cited by 2
“[19] designed an adaptive DNN splitting algorithm, which can find the optimal splitting strategy under dynamic and time-varying network conditions.”
MethodSystem support and mechanisms for adaptive edge-to-cloud DNN model serving
Reisinger, M., Frangoudis, P. A., & Dustdar, S. (2021) · IEEE International Conference on Cloud Engineering · doi:10.1109/ic2e52221.2021.00046 · Cited by 2
“The state of the art either allows a single DNN split point at a time and does not explicitly consider early-exit architectures [2]–[4], or assumes the full model is available both on-device and on a remote server [2], [5], or does not focus on system support for DNN orchestration [3], [6]–[8].”
BackgroundKilcioglu, E., Stupia, I., & Vandendorpe, L. (2023) · IEEE Access · doi:10.1109/access.2023.3314381 · Cited by 2
“There are various solutions proposed, but deep neural network (DNN) partitioning is considered to be an effective solution for workload distribution in such applications [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], [31], [32], [33], [34], [35], [36].”
BackgroundPreventing Information Leakage with Neural Architecture Search.Preprint
Zhang, S., Xiang, L., Li, C., Wang, Y., Liu, Z., Zhang, Q., & Li, B. (2019) · arXiv (Cornell University) · Cited by 2
Distributed deep learning inference in fog networks
Chowdhury, A. G. (2020) · Aaltodoc (Aalto University) · Cited by 2
Hybrid-Parallel: Achieving High Performance and Energy Efficient Distributed Inference on RobotsPreprint
Sun, Z., Guan, X., Wang, J., Song, H., Qing, Y., Shen, T., Huang, D., Liu, F., & Cui, H. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2405.19257 · Cited by 2
“…on the fact that the amounts of output data in some intermediate layers of a DNN model are significantly smaller than that of its raw input data [17], DNN layer partitioning strategies constitute various trade-offs between computation and transmission, taking into account application-specific…”
BackgroundMethodSelf-citationData-aware Dynamic Execution of Irregular Workloads on Heterogeneous SystemsPreprint
Bai, Z., Wu, D., Dangi, P., Wijerathne, D., Miriyala, V. P. K., & Mitra, T. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2502.06304 · Cited by 2
The Limitations of Deep Learning Methods in Realistic Adversarial Settings
Kaya, Y. (2023) · University Libraries (University of Maryland) · doi:10.13016/dspace/9fyt-k9dh · Cited by 2
Ask the Expert: Collaborative Inference for Vision Transformers with Near-Edge AcceleratorsPreprint
Liu, H., & Fahmy, S. A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.13334 · Cited by 2
Ye, C., Zhang, R., Li, X., Deng, W., Wang, J., & Shao, S. (2025) · Information · doi:10.3390/info16121091 · Cited by 1
“Some approaches rely on specialized graph-based heuristics, like those in DADS [25] and QDMP [26], to accelerate the search process for real-time inference.”
Jeong, H.-J., Boo, H., Bae, J., Jeon, M., & Huh, E. (2024) · International Conference on Intelligent Information Technology (ICIIT) · doi:10.1145/3654522.3654596 · Cited by 1
“Research endeavors like DADS [6] have explored employing additional CPU resources for inference.”
Mhasakar, P. M., Doshi, K., Wang, N., Ho, S.-S., & Ling, H. (2023) · ACM/IEEE Symposium on Edge Computing (SEC) · doi:10.1145/3583740.3626810 · Cited by 1
“Papers like Neurosurgeon [5], Deep Things [6], DADS [7], QDMP [8], Auto-Split [9], DNN schedule [10], etc., have proposed different splitting/distribution techniques to perform this edge cloud partitioning like layer-based splitting, graph-based splitting, fused tile partitioning etc.”
MethodCo-Inference Over Wireless Multi - Hop Networks
Zhang, X., Lin, Z., Zhou, C., Yang, J., & Liu, Y. (2025) · IEEE Wireless Communications and Networking Conference · doi:10.1109/wcnc61545.2025.10978661 · Cited by 1
Otimizando Saídas Antecipadas em Redes Neurais Profundas: Como Lidar com Buffers?
Pacheco, R. G., Bajpai, D. J., Shifrin, M., Couto, R. S., Menasché, D. S., Hanawal, M. K., & Campista, M. E. M. (2025) · doi:10.5753/sbrc.2025.6319 · Cited by 1
MPLS: Stacking Diverse Layers Into One Model for Decentralized Federated Learning
Xu, Y., Yao, Z., Xu, H., Liao, Y., & Xie, Z. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-99854-6_13 · Cited by 1
Ng, N., Irwin, D., Swami, A., Towsley, D., & Shenoy, P. J. (2026) · ACM/SPEC International Conference on Performance Engineering (ICPE) · doi:10.1145/3777884.3797816 · Cited by 1
“…are governed by independent Poisson processes, and the composition of independent Poisson processes results in a new Poisson process [37], it follows that the aggregate edge workload is also Poisson with an arrival rate λ edge = ∑ mi = 0 λ i , where λ i denotes the workload at the i -th device.”
Enabling Efficient Synergistic Multi-view Inference Across Heterogeneous Edge Devices
Dong, F., Chen, R., Fu, S., Cheng, W., Zhou, R., & Zhang, X. (2026) · ACM Transactions on Sensor Networks · doi:10.1145/3785358 · Cited by 1
“To tackle these performance bottlenecks, edge intelligence ( EI ) has emerged [1, 30, 44], leveraging edge computation together with model partitioning [13, 16] to reduce both computation and communication loads.”
ELM-Split: Energy-Latency-Memory Aware Splitting of DNN Models in Edge-Cloud Ecosystem
Saxena, R., Saha, R. K., Singhal, R., & Nambiar, M. K. (2025) · International Conference on AI-ML Systems (AIMLSystems) · doi:10.1109/aimlsystems67835.2025.11330640 · Cited by 1
“Hu et al. [4] proposed a dynamic adaptive DNN surgery scheme, optimizing inference by adaptively splitting computation between the edge and cloud based on network conditions to minimize latency and maximize throughput.”
Accelerating CNN Inference via In-Network Computing in Information-Centric Networking
Hu, K., Deng, H., & Ma, B. (2026) · Electronics · doi:10.3390/electronics15040775 · Cited by 1
Intelligent O-RAN Optimization: AI/ML-Enabled Dynamic Prediction for Adaptive Rate Control
Kumar, S. (2026) · Journal of Network and Systems Management · doi:10.1007/s10922-025-10016-3 · Cited by 1
“[ 19 ] provided a foundational survey on the integration of machine learning into self-organizing networks (SON), outlining supervised and reinforce - ment learning techniques for dynamic rate and link adaptation across cellular net - works.”
Hao, Y., Ding, N., & Gao, H. (2026) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2026.114346 · Cited by 1
Reinforcement Learning Based Edge-Assisted Dynamic Inference for Mobile Vision
Liu, Y., Xiao, L., Li, T., Lv, Z., Qiao, Z., & Xiong, H. (2025) · Globecom · doi:10.1109/globecom59602.2025.11431768 · Cited by 1
“The input image is represented by the source vertex v 0 , the intermediate activations are represented by intermediate vertices and the inference result is represented by the sink vertex v V , in which V is the number of data representations [13].”
DMSF: A Dynamic Model Splitting Framework for Edge-Cloud Collaborative Inference
Zhang, X., Wang, L., Wu, X., Xu, L., Hou, Y., Jiang, Y., & Fei, A. (2025) · Globecom · doi:10.1109/globecom59602.2025.11432620 · Cited by 1
“While early methods [6], [7] focused on simple single-branch models with clear layer-wise structures, others [8] extended to multi-branch models like InceptionNet [9] by applying graph-cut techniques, leveraging the relative independence of their branches.”
Optimization Approaches for Distributed AI Models on Edge Devices
Fernandez, N., Amurrio, A., & Vaerenbergh, S. V. (2025) · Springer eBooks · doi:10.1007/978-981-95-0351-3_5 · Cited by 1
ApproxBit: Efficient Video Analytics through Latency-Aware Offloading with Learned Binary Codes
Kim, H., Prasanna, S., Li, Y., Chaterji, S., & Bagchi, S. (2026) · doi:10.1145/3774906.3802757 · Cited by 1
“Many existing solutions [2, 11, 20, 28, 32, 35, 41, 69] follow a strategy that balances high accuracy with reasonable latency.”
ScalPipe: Scalable Collaborative Pipeline Inference for Distributed Heterogeneous Devices
Wang, N., Shi, G., Wang, W., Zhong, X., Liu, J., & Li, Z. (2026) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2026.3693694 · Cited by 1
“In pipeline collaborative inference [10], [11], [12], the DNN is partitioned vertically across layers, and the resulting layer segments are allocated to different devices for execution in a pipeline manner.”
Uncertainty-Guided Adaptive Model Hopping for Efficient Object Tracking on Edge Devices
Kumar, D., Rios, A., Ahuja, N., Tickoo, O., & Trivedi, A. R. (2026) · International Symposium on Quality Electronic Design (ISQED) · doi:10.1109/isqed69900.2026.11534770 · Cited by 1
“DADS [1] performs dynamic DNN partitioning be-tween edge and cloud based on network conditions to reduce latency.”
SAIL: Redesigning Collaborative Language Inference with a Single Server-to-Mobile Handoff
Park, G., Han, S., Cho, Y., Park, C., & Lee, K. (2026) · International Conference on Mobile Systems, Applications, and Services · doi:10.1145/3745756.3809224 · Cited by 1
LAUA: Handling Missing Modalities and Unpaired Data in Multimodal Federated Learning
Wei, Y., Zhou, X., Feng, S., Hu, C., Yan, X., & Jiang, J. (2026) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3770855.3817772 · Cited by 1
Self-citationIntroduzindo a qualidade de imagem como uma nova condição de particionamento de DNN na borda
Pacheco, R. G., & Couto, R. S. (2020) · doi:10.5753/sbrc.2020.12272 · Cited by 1
“De forma geral, a Figura 3 mostra que o tipo de rede e sua respectiva taxa de envio representam um fator decisivo para determinar a estratégia de processamento adotada [Hu et al., 2019].”
BackgroundInfluentialMartín, C., Llopis, L., Rubio, B., Díáz, M., & Díaz-Yáñez, F. (2021) · Servizo de Publicacións da UDC eBooks · doi:10.17979/spudc.9788497498043.744 · Cited by 1
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.