MoDNN: Local distributed mobile computing system for Deep Neural Network
The DOI matches its Crossref record; 19 of the 20 references it cites were found in a registry or an index.
Abstract
Although Deep Neural Networks (DNN) are ubiquitously utilized in many applications, it is generally difficult to deploy DNNs on resource-constrained devices, e.g., mobile platforms. Some existing attempts mainly focus on client-server computing paradigm or DNN model compression, which require either infrastructure supports or special training phases, respectively. In this work, we propose MoDNN - a local distributed mobile computing system for DNN applications. MoDNN can partition already trained DNN models onto several mobile devices to accelerate DNN computations by alleviating device-level computing cost and memory usage. TWo model partition schemes are also designed to minimize non-parallel data delivery time, including both wakeup time and transmission time. Experimental results show that when the number of worker nodes increases from 2 to 4, MoDNN can accelerate the DNN computation by 2.17-4.28 X. Besides the parallel execution, the performance speedup also partially comes from the reduction of the data delivery time, e.g., 30.02% w.r.t. conventional 2D-grids partition.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“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.”
From Convergence of Edge Computing and Deep Learning: A Comprehensive Survey (2020) · cites this paper as [177]
In the citing paper
“The experiment shows that with two to four worker nodes, MoDNN accelerates the DNN model inference by 2.17-4.28×.”
From Edge Intelligence: Paving the Last Mile of Artificial Intelligence With Edge Computing (2019) · cites this paper as [85]
Citations
Cites
Very Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148 · cited 4×
“A representative example is V GG [1], which demonstrates state-of-the-art performance in ImageNet Large Scale Visual Recognition Challenge 2014 (ILSVRC14).”
BackgroundMethodInfluentialGradient-based learning applied to document recognition
Lecun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998) · Proceedings of the IEEE · doi:10.1109/5.726791 · Cited by 58,805 · cited 1×
“1 depicts our measurement results of the computing time and the memory usage of different network layers from three popular DNN models running on smartphones – Lenet [9], Inception - BN [10], and V GG [1].”
BackgroundBatch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Ioffe, S., & Szegedy, C. (2015) · arXiv · doi:10.48550/arxiv.1502.03167 · Cited by 27,818 · cited 1×
“1 depicts our measurement results of the computing time and the memory usage of different network layers from three popular DNN models running on smartphones – Lenet [9], Inception - BN [10], and V GG [1].”
BackgroundMapReduce: simplified data processing on large clusters
Dean, J., & Ghemawat, S. (2008) · Communications of the ACM · doi:10.1145/1327452.1327492 · Cited by 18,476 · cited 1×
“MapReduce is a programming model for simplifying parallel data processing in distributed systems [7].”
BackgroundBatch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Ioffe, S. (2024) · arXiv (Cornell University) · doi:10.57702/o9raffed · Cited by 14,214
A tutorial on spectral clustering
von Luxburg, U. (2007) · Statistics and Computing · doi:10.1007/s11222-007-9033-z · Cited by 10,349
Some simplified NP-complete graph problems
Garey, M. R., Johnson, D. S., & Stockmeyer, L. (1976) · Theoretical Computer Science · doi:10.1016/0304-3975(76)90059-1 · Cited by 2,059 · cited 1×
“If we consider the weight matrix as an undirected graph, generating k clusters with minimal connections between them is a NP hard problem [14].”
BackgroundMXNet: A Flexible and Efficient Machine Learning Library for Heterogeneous Distributed Systems
Chen, T., Li, M., Li, Y., Lin, M., Wang, N., Wang, M., Xiao, T., Xu, B., Zhang, C., & Zhang, Z. (2015) · arXiv · doi:10.48550/arxiv.1512.01274 · Cited by 1,819 · cited 2×
“We modified and recompiled the MXNet libraries so that it can support Android systems with ARM architecture through JAVA Native Interface (JNI) [17].”
MethodCo-clustering documents and words using bipartite spectral graph partitioning
Dhillon, I. S. (2001) · Proceedings of the seventh ACM SIGKDD international conference on Knowledge discovery and data mining · doi:10.1145/502512.502550 · Cited by 1,697 · cited 1×
“Therefore, spectral co-clustering algorithm is introduced to address this drawback by normalizing the original connection matrix A to A norm and performing Singular Value Decomposition (SVD) on A norm [16].”
MethodScaling distributed machine learning with the parameter server
Li, M., Andersen, D. G., Park, J. W., Smola, A. J., Ahmed, A. A., Josifovski, V., Long, J. E., Shekita, E. J., & Su, B.-Y. (2014) · Operating Systems Design and Implementation · doi:10.5555/2685048.2685095 · Cited by 1,021
Research works printed but not linked: 2
- han, 2015, Deep compression: Compressing deep neural network with pruning, trained quantization and huffman coding, Computing ResearchRepository, 2
- 0, Wifi Alliance Wifi Direct Specifications
Cited by
Items 301–327 of 327
Master-Device-Coordinated Optimization for Distributed CNN Inference at the Edge
Guo, H. Y., & Wang, X. (2025) · International Symposium on Robotics, Artificial Intelligence and Information Engineering (RAIIE) · doi:10.1109/raiie65740.2025.11140395
“MoDNN [10] adopts a layer-wise parallelization strategy for CNN inference, requiring partitioning and immediate merging of each convolutional layer, which results in prohibitive communication costs.”
Adaptive partitioning of DNNs for resource-efficient inference on edge clusters
Nazir, A., Lone, F. R., & Najar, A. A. (2025) · Cluster Computing · doi:10.1007/s10586-025-05386-x · Cited by 1
“MoDNN utilizes the Sleep and Wake times of mobile devices as constraints besides communication overhead for distribution of neural network in a wireless environment [34].”
Luo, Y., Qi, J., Li, L., Chen, R., Wu, X., Li, C., & Zhao, C. (2025) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2025.3619775
“Although the previousworks[11],[12],[13],[14]relatedtolocalcollaborative inferenceoptimizetheperformance,theydonotconsiderthe congestiondelayinMLCCI.”
InfluentialDistributed Convolutional Neural Network Training on Mobile and Edge Clusters
Rama, P., Threadgill, M., & Gerstlauer, A. (2025) · IFIP advances in information and communication technology · doi:10.1007/978-3-032-07102-6_6 · Cited by 1
Zhai, H., Wang, Y., Peng, X., Li, L., & Li, X. (2025) · Lecture notes in computer science · doi:10.1007/978-3-032-10459-5_36
Security Robustness Evolution of Split Edge-Cloud Neural NetworksPreprint
DOUCH, S., ABID, M. R., ZINE-DINE, K., BOUZIDI, D., & BENHADDOU, D. (2023) · Research Square · doi:10.21203/rs.3.rs-3304497/v1
PIDC: Padding-Aware IoT Device Collaboration for Accelerating DNN Inference
Jiang, W., Han, H., Qian, L. P., Wei, F., Qin, S., & Feng, G. (2025) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2025.3633269 · Cited by 1
“(cid:2) MoDNN (MD) [47]: This approach allocates computation tasks among devices based on their computing capabilities, but ignores communication overhead.”
A Security Analysis of CNN Partitioning Strategies for Distributed Inference at the Edge
Mehrafrooz, F., Siyadatzadeh, R., Mentens, N., & Stefanov, T. P. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-16342-4_15
“…have proposed approaches to distribute and optimize CNN model inference across multiple edge devices to improve the overall system performance [7–14], to reduce memory requirements [7–9,11, 13, 15 ] and energy consumption [8,11, The main focus of the aforementioned studies has been on…”
Li, Z., & Abdullah, S. N. H. S. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.6318568
PowerShift: Leveraging Power-Aware Weight Approximations for Neural Network Acceleration
Mishra, V., Mittal, S., & Chatterjee, U. (2026) · International Conference on VLSI Design · doi:10.1109/vlsid68508.2026.00079
“However, deploying DNNs on resource-constrained edge devices poses a significant challenge due to the high power consumption associated with multiply-and-accumulate (MAC) operations [2], [3].”
Sun, B., Teng, Y., & Wang, N. (2026) · Lecture notes in computer science · doi:10.1007/978-981-95-8405-5_15
Joshi, R., Ghosh, J., Tanna, R., Kumar, R., Praveenlal, E., & Kumar, S. (2026) · IEEE Transactions on Plasma Science · doi:10.1109/tps.2026.3684924
“Edge computing deployments are evaluated for potential distributed inference architectures [35], [36], [37].”
Acceleration Strategy for Edge-Assisted Object Recognition towards Internet of Vehicles
Liu, Z., Ma, R., Wu, C., Zhang, X., Fan, H., & Zhu, Z. (2026) · Lecture notes in electrical engineering · doi:10.1007/978-981-95-6860-4_53
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
“MoDNN [35] splits convolutional layers and allocates work according to device capacity.”
Wang, Z., Ren, A., Liu, D., Li, W., Fang, H., Song, Y., Tan, Y., Zhong, K., & Qin, C. (2026) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2026.3693876
“Traditional sequential execution approaches [4], [10] transmit substantial volumes of intermediate feature maps between devices, resulting in significant communication overhead, as illustrated in Fig.”
InfluentialLate Breaking Results: Adaptive Ensembles of Dynamic DNNs for Collaborative Edge Inference
Hu, M., Singh, A. K., Hare, J. S., & Merrett, G. V. (2026) · Design, Automation, and Test in Europe · doi:10.23919/date69613.2026.11539467
“Distributed inference methods parallelize computation across devices, but layer-wise dependencies induce substantial intermediate communication, making performance sensitive to network fluctuations and requiring complex scheduling [3]– [5].”
ADaPS-FC: Enhancing Performance Through Channel Division on Edge Computing Platforms
Cuadrat, J. M., & Egger, B. (2026) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3820058
FlexPie: Accelerate Distributed Inference on Edge Devices with Flexible Combinatorial Optimization
Zhang, R., Jiang, H., Geng, J., Ma, Y., Zhu, C., & Wang, H. H. (2026) · Lecture notes in computer science · doi:10.1007/978-981-95-3827-0_2
EdgeSpec: Distributed Speculative Decoding for Large Language Models at Edge
Chen, Y., Tian, M., Qiu, C., Wang, X. L., Yao, H., He, Y., & Liu, Y. (2026) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom59046.2026.11571680
“Early systems like Neu-rosurgeon [35] proposed dynamic DNN partitioning between edge and cloud, while DeepThings [36] and MoDNN [37] extended this to multi nodes execution.”
Self-citationMyroshnyk, Y., & Biriukov, Y. (2026) · IEEE Access · doi:10.1109/access.2026.3719609
“MoDNN [9] distributes inference across multiple mobile devices using a work-stealing scheduler.”
Tian, X., Mao, X., & Zhou, X. (2026) · The Journal of Supercomputing · doi:10.1007/s11227-026-08777-y
“Model partitioning divides a DNN into a local part and an edge server part, allowing the UE to process the initial layers and then offload intermediate features to the edge server [12].”
Communication-Budgeted Chunk-Aware Mixed Model Parallelism for Collaborative Inference
Chen, Y., Qiu, C., Wang, X., Liu, Z., Zhao, Y., Wang, Y., & Luo, T. (2026) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2026.3723098
“A common line of work reduces communication volume through architectural or execution shortcuts, such as early-exit inference and selective transmission of intermediate activations [35], [36].”
Self-citationPractical Modeling for Split DNN Inference on Near-Edge Accelerators
Liu, H., Fouda, M. E., Eltawil, A. M., & Fahmy, S. A. (2026) · IEEE International Conference on Edge Computing and Communications (EDGE) · doi:10.1109/edge72783.2026.00011
“Network wake-up time and transmission time were considered in [29], but without a general formulation.”
Distributed CNN inference system based on aircomp
Liu, Q., Du, X., Tian, C., Hu, H., & Wang, J. (2026) · Wireless Networks · doi:10.1007/s11276-026-04216-1
“For example, distributed CNN inference on multiple edge devices in a parallel fashion has been studied in [9–12 ], which demonstrates its benefits in reducing infer - ence latency, avoiding memory overflow on devices, and utilizing idle computing resources [13].”
Zhao, W., & Zeng, J. (2026) · Discover Artificial Intelligence · doi:10.1007/s44163-026-02208-w
“Existing cloud-edge intelligence and model-offloading systems mainly focus on computation partition, feature placement, or inference scheduling across resource-constrained devices [4–6].”
Wu, Y., Siyan, Z., Luo, F., Zhu, X., Wen, J., Li, Y., Liu, W., & Wu, X. (2026) · International Conference on Artificial Intelligence, Internet of Things and Cloud Computing Technology (AIoTC) · doi:10.1109/aiotc71250.2026.11689027
“MoDNN [4] relies on spatial partitioning of input feature maps on convolutional layers to achieve multi-device parallelism.”
Wang, J., Liao, X., Liu, X., Li, Y., Huo, Z., Suo, J., Shen, R., Xie, X., Xu, X., Wang, L., et al. (2026) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2026.3738731
“Geo-distributed Training: MoDNN [75] and CoEdge [76] divide DNN workloads dynamically based on device computing capabilities, allowing for collaboration with edges.”
Metadata
- Authors
- Mao, J., Chen, X., Nixon, K. W., Krieger, C., & Chen, Y.
- Affiliations
- University of Pittsburgh (US)
- George Mason University (US)
- University of Maryland, College Park (US)
- Venue
- Design, Automation & Test in Europe Conference & Exhibition (DATE), 2017
- Publisher
- IEEE
- Published
- 2017-03
- Pages
- 1396-1401
- OpenAlex
- W2612193523
- Cited by
- 331
- References
- 20
- Citations by year
- 2026: 21 · 2025: 35 · 2024: 47 · 2023: 62 · 2022: 43 · 2021: 51 · 2020: 38 · 2019: 17 · 2018: 11 · 2017: 6
- FWCI
- 15.69
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- Design, Automation, and Test in Europe (publishedVersion)
- Funders
- National Science Foundation
- Topics
- Physical Sciences › Computer Science › Computer Networks and Communications › IoT and Edge/Fog Computing
- Physical Sciences › Computer Science › Hardware and Architecture › Parallel Computing and Optimization Techniques
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Keywords
- deep neural network, mobile computing, distributed computing, model partitioning, resource-constrained devices, computational acceleration, parallel execution, memory reduction, worker nodes
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.