mlsanityData to 14 Oct 2026

Split Computing and Early Exiting for Deep Learning Applications: Survey and Research Challenges

arXiv 2103.04505 ↗First posted 8 Mar 2021eess.SP cs.LGcited by 115

On the map: Efficient LLM inference and serving → Early exit and edge inference

Abstract (arXiv)

Mobile devices such as smartphones and autonomous vehicles increasingly rely on deep neural networks (DNNs) to execute complex inference tasks such as image classification and speech recognition, among others. However, continuously executing the entire DNN on mobile devices can quickly deplete their battery. Although task offloading to cloud/edge servers may decrease the mobile device's computational burden, erratic patterns in channel quality, network, and edge server load can lead to a significant delay in task execution. Recently, approaches based on split computing (SC) have been proposed, where the DNN is split into a head and a tail model, executed respectively on the mobile device and on the edge server. Ultimately, this may reduce bandwidth usage as well as energy consumption. Another approach, called early exiting (EE), trains models to embed multiple "exits" earlier in the architecture, each providing increasingly higher target accuracy. Therefore, the trade-off between accuracy and delay can be tuned according to the current conditions or application demands. In this paper, we provide a comprehensive survey of the state of the art in SC and EE strategies by presenting a comparison of the most relevant approaches. We conclude the paper by providing a set of compelling research challenges.

Key facts. Cited by 115 papers in the map (data through 14 Oct 2026). Placed in Early exit and edge inference, one of 12 directions in Efficient LLM inference and serving. Builds on 87 of its 183 references that are in the map. First posted on arXiv 8 Mar 2021.

Builds on

87 of 183 references.

References · 87

  1. 2017Attention Is All You Needcited by 60,361
  2. 2015Deep Residual Learning for Image Recognitioncited by 52,096
  3. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,399
  4. 2014Adam: A Method for Stochastic Optimizationcited by 37,240
  5. 2014Microsoft COCO: Common Objects in Contextcited by 17,608
  6. 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,364
  7. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,372
  8. 2015Distilling the Knowledge in a Neural Networkcited by 10,324
  9. 2015Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networkscited by 9,639
  10. 2014ImageNet Large Scale Visual Recognition Challengecited by 8,498
  11. 2017Mask R-CNNcited by 7,277
  12. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 7,005
38 more
  1. 2016Densely Connected Convolutional Networkscited by 6,720
  2. 2015Rethinking the Inception Architecture for Computer Visioncited by 5,475
  3. 2016Feature Pyramid Networks for Object Detectioncited by 5,036
  4. 2018GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understandingcited by 4,899
  5. 2018MobileNetV2: Inverted Residuals and Linear Bottleneckscited by 4,795
  6. 2014Going Deeper with Convolutionscited by 4,241
  7. 2016SQuAD: 100,000+ Questions for Machine Comprehension of Textcited by 4,148
  8. 2017MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applicationscited by 3,845
  9. 2013Playing Atari with Deep Reinforcement Learningcited by 3,274
  10. 2016Wide Residual Networkscited by 3,070
  11. 2017Prototypical Networks for Few-shot Learningcited by 3,015
  12. 2019ALBERT: A Lite BERT for Self-supervised Learning of Language Representationscited by 2,790
  13. 2017A Broad-Coverage Challenge Corpus for Sentence Understanding through Inferencecited by 2,682
  14. 2018YOLOv3: An Incremental Improvementcited by 2,659
  15. 2015Character-level Convolutional Networks for Text Classificationcited by 2,330
  16. 2015Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Codingcited by 2,329
  17. 2017Rethinking Atrous Convolution for Semantic Image Segmentationcited by 2,161
  18. 2015Learning both Weights and Connections for Efficient Neural Networkscited by 2,031
  19. 2019Searching for MobileNetV3cited by 1,994
  20. 2016Adversarial examples in the physical worldcited by 1,769
  21. 2016MS MARCO: A Human Generated MAchine Reading COmprehension Datasetcited by 1,605
  22. 2017Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inferencecited by 1,593
  23. 2016Neural Architecture Search with Reinforcement Learningcited by 1,499
  24. 2016YOLO9000: Better, Faster, Strongercited by 1,431
  25. 2016Pruning Filters for Efficient ConvNetscited by 1,143
  26. 2017Learning Transferable Architectures for Scalable Image Recognitioncited by 1,101
  27. 2019EfficientDet: Scalable and Efficient Object Detectioncited by 1,020
  28. 2017SemEval-2017 Task 1: Semantic Textual Similarity - Multilingual and Cross-lingual Focused Evaluationcited by 1,001
  29. 2020Designing Network Design Spacescited by 960
  30. 2018Neural Network Acceptability Judgmentscited by 958
  31. 2018MnasNet: Platform-Aware Neural Architecture Search for Mobilecited by 943
  32. 2018Variational image compression with a scale hyperpriorcited by 935
  33. 2015Deep Learning and the Information Bottleneck Principlecited by 802
  34. 2016Sequence-Level Knowledge Distillationcited by 801
  35. 2017Channel Pruning for Accelerating Very Deep Neural Networkscited by 702
  36. 2016End-to-end Optimized Image Compressioncited by 696
  37. 2019Is BERT Really Robust? A Strong Baseline for Natural Language Attack on Text Classification and Entailmentcited by 671
  38. 2013Network In Networkcited by 605

50 most cited of 87 All 87 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2021: 6; 2022: 12; 2023: 16; 2024: 28; 2025: 28; 2026: 25.

115 papers.

Sort:

By area:

  1. 2024Mobile Edge Intelligence for Large Language Models: A Contemporary Surveycited by 128
  2. 2021Adaptive Inference through Early-Exit Networks: Design, Challenges and Directionscited by 75
  3. 2023Privacy and Efficiency of Communications in Federated Split Learningcited by 21
  4. 2022Dynamic Split Computing for Efficient Deep Edge Intelligencecited by 18
  5. 2023Towards Anytime Classification in Early-Exit Architectures by Enforcing Conditional Monotonicitycited by 16
  6. 2022SC2 Benchmark: Supervised Compression for Split Computingcited by 15
  7. 2022SplitNets: Designing Neural Architectures for Efficient Distributed Computing on Head-Mounted Systemscited by 14
  8. 2025Adaptive Semantic Token Communication for Transformer-based Edge Inferencecited by 14
  9. 2023SplitEE: Early Exit in Deep Neural Networks with Split Computingcited by 13
  10. 2023FrankenSplit: Efficient Neural Feature Compression with Shallow Variational Bottleneck Injection for Mobile Edge Computingcited by 12
  11. 2022Binarizing Split Learning for Data Privacy Enhancement and Computation Reductioncited by 11
  12. 2022Resource-Constrained Edge AI with Early Exit Predictioncited by 11
38 more
  1. 2024Conditional computation in neural networks: principles and research trendscited by 10
  2. 2025Towards Understanding How Knowledge Evolves in Large Vision-Language Modelscited by 10
  3. 2021Resource-Efficient Deep Learning: A Survey on Model-, Arithmetic-, and Implementation-Level Techniquescited by 9
  4. 2022Efficient High-Resolution Deep Learning: A Surveycited by 9
  5. 2024I-SplitEE: Image classification in Split Computing DNNs with Early Exitscited by 9
  6. 2021Privacy Vulnerability of Split Computing to Data-Free Model Inversion Attackscited by 8
  7. 2022I-SPLIT: Deep Network Interpretability for Split Computingcited by 7
  8. 2022Energy Consumption of Neural Networks on NVIDIA Edge Boards: an Empirical Modelcited by 7
  9. 2022Update Compression for Deep Neural Networks on the Edgecited by 6
  10. 2023Lambda-Split: A Privacy-Preserving Split Computing Framework for Cloud-Powered Generative AIcited by 6
  11. 2024FOOL: Addressing the Downlink Bottleneck in Satellite Computing with Neural Feature Compressioncited by 6
  12. 2025A Multi-task Supervised Compression Model for Split Computingcited by 6
  13. 2024Cross-layer Attention Sharing for Pre-trained Large Language Modelscited by 5
  14. 2024Semantic Edge Computing and Semantic Communications in 6G Networks: A Unifying Survey and Research Challengescited by 5
  15. 2024Unlocking Visual Secrets: Inverting Features with Diffusion Priors for Image Reconstructioncited by 5
  16. 2025A Survey on Dynamic Neural Networks: from Computer Vision to Multi-modal Sensor Fusioncited by 5
  17. 2025DSD: A Distributed Speculative Decoding Solution for Edge-Cloud Agile Large Model Servingcited by 5
  18. 2024Dynamic Semantic Compression for CNN Inference in Multi-access Edge Computing: A Graph Reinforcement Learning-based Autoencodercited by 4
  19. 2025AVERY: Intent-Driven Adaptive VLM Split Computing via Embodied Self-Awareness for Efficient Disaster Response Systemscited by 4
  20. 2026ViTNT-FIQA: Training-Free Face Image Quality Assessment with Vision Transformerscited by 4
  21. 2022Neural Architecture Search for Improving Latency-Accuracy Trade-off in Split Computingcited by 3
  22. 2023Anticipate, Ensemble and Prune: Improving Convolutional Neural Networks via Aggregated Early Exitscited by 3
  23. 2023Split-Et-Impera: A Framework for the Design of Distributed Deep Learning Applicationscited by 3
  24. 2023ATHEENA: A Toolflow for Hardware Early-Exit Network Automationcited by 3
  25. 2023Slimmable Encoders for Flexible Split DNNs in Bandwidth and Resource Constrained IoT Systemscited by 3
  26. 2023Salted Inference: Enhancing Privacy while Maintaining Efficiency of Split Inference in Mobile Computingcited by 3
  27. 2023Adaptive Early Exiting for Collaborative Inference over Noisy Wireless Channelscited by 3
  28. 2024On the Impact of Black-box Deployment Strategies for Edge AI on Latency and Model Performancecited by 3
  29. 2024LightFF: Lightweight Inference for Forward-Forward Algorithmcited by 3
  30. 2024Distributed Inference on Mobile Edge and Cloud: An Early Exit based Clustering Approachcited by 3
  31. 2024SpikeBottleNet: Spike-Driven Feature Compression Architecture for Edge-Cloud Co-Inferencecited by 3
  32. 2025DRAG: Data Reconstruction Attack using Guided Diffusioncited by 3
  33. 2025Memory- and Latency-Constrained Inference of Large Language Models via Adaptive Split Computingcited by 3
  34. 2026FlexSpec: Frozen Drafts Meet Evolving Targets in Edge-Cloud Collaborative LLM Speculative Decodingcited by 3
  35. 2023Architectural Vision for Quantum Computing in the Edge-Cloud Continuumcited by 2
  36. 2023Sustainable Edge Intelligence Through Energy-Aware Early Exitingcited by 2
  37. 2023Adversarial Attacks to Latent Representations of Distributed Neural Networks in Split Computingcited by 2
  38. 2024Multimodal Adaptive Inference for Document Image Classification with Anytime Early Exitingcited by 2

50 most cited of 115 All 115 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2022Enable Deep Learning on Mobile Devices: Methods, Systems, and Applications
  2. 2021ESAI: Efficient Split Artificial Intelligence via Early Exiting Using Neural Architecture Search
  3. 2025DistrEE: Distributed Early Exit of Deep Neural Network Inference on Edge Devices
  4. 2023AdaMEC: Towards a Context-Adaptive and Dynamically-Combinable DNN Deployment Framework for Mobile Edge Computing
  5. 2022Fluid Batching: Exit-Aware Preemptive Serving of Early-Exit Neural Networks on Edge NPUs
  6. 2023Adaptive DNN Surgery for Selfish Inference Acceleration with On-demand Edge Resource
  7. 2024Early-Exit meets Model-Distributed Inference at Edge Networks
  8. 2022Predictive Exit: Prediction of Fine-Grained Early Exits for Computation- and Energy-Efficient Inference