mlsanityData to 7 Oct 2026

Why should we add early exits to neural networks?

arXiv 2004.12814 ↗First posted 27 Apr 2020cs.NE cs.LG stat.MLcited by 70

Not on the map. Why?

Cited by 23 in Efficient LLM inference and serving, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

Deep neural networks are generally designed as a stack of differentiable layers, in which a prediction is obtained only after running the full stack. Recently, some contributions have proposed techniques to endow the networks with early exits, allowing to obtain predictions at intermediate points of the stack. These multi-output networks have a number of advantages, including: (i) significant reductions of the inference time, (ii) reduced tendency to overfitting and vanishing gradients, and (iii) capability of being distributed over multi-tier computation platforms. In addition, they connect to the wider themes of biological plausibility and layered cognitive reasoning. In this paper, we provide a comprehensive introduction to this family of neural networks, by describing in a unified fashion the way these architectures can be designed, trained, and actually deployed in time-constrained scenarios. We also describe in-depth their application scenarios in 5G and Fog computing environments, as long as some of the open research questions connected to them.

Key facts. Cited by 70 papers in the map (data through 7 Oct 2026). First posted on arXiv 27 Apr 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 8; 2022: 14; 2023: 17; 2024: 13; 2025: 14; 2026: 4.

70 papers.

Sort:

By area:

  1. 2022Fast Inference from Transformers via Speculative Decodingcited by 1,615
  2. 2025Universal Model Routing for Efficient LLM Inferencecited by 80
  3. 2023EE-LLM: Large-Scale Training and Inference of Early-Exit Large Language Models with 3D Parallelismcited by 69
  4. 2023When Does Confidence-Based Cascade Deferral Suffice?cited by 58
  5. 2021Zero Time Waste: Recycling Predictions in Early Exit Neural Networkscited by 39
  6. 2021Multi-Exit Vision Transformer for Dynamic Inferencecited by 28
  7. 2023The Benefits of Bad Advice: Autocontrastive Decoding across Model Layerscited by 28
  8. 2021ProgFed: Effective, Communication, and Computation Efficient Federated Learning by Progressive Trainingcited by 27
  9. 2024Fast yet Safe: Early-Exiting with Risk Controlcited by 27
  10. 2021Single-Layer Vision Transformers for More Accurate Early Exits with Less Overheadcited by 26
  11. 2022Predictive Exit: Prediction of Fine-Grained Early Exits for Computation- and Energy-Efficient Inferencecited by 24
  12. 2023Fixing Overconfidence in Dynamic Neural Networkscited by 19
38 more
  1. 2022Dynamic Split Computing for Efficient Deep Edge Intelligencecited by 18
  2. 2024HADES: Hardware Accelerated Decoding for Efficient Speculation in Large Language Modelscited by 15
  3. 2024EE-Tuning: An Economical yet Scalable Solution for Tuning Early-Exit Large Language Modelscited by 13
  4. 2022OccamNets: Mitigating Dataset Bias by Favoring Simpler Hypothesescited by 11
  5. 2022Unsupervised Early Exit in DNNs with Multiple Exitscited by 10
  6. 2024Conditional computation in neural networks: principles and research trendscited by 10
  7. 2021Improving the Accuracy of Early Exits in Multi-Exit Architectures via Curriculum Learningcited by 9
  8. 2022Efficient High-Resolution Deep Learning: A Surveycited by 9
  9. 2024NACHOS: Neural Architecture Search for Hardware Constrained Early Exit Neural Networkscited by 9
  10. 2021AHAR: Adaptive CNN for Energy-efficient Human Activity Recognition in Low-power Edge Devicescited by 8
  11. 2023Spiking Neural Networks and Bio-Inspired Supervised Deep Learning: A Surveycited by 8
  12. 2021Improving Anytime Prediction with Parallel Cascaded Networks and a Temporal-Difference Losscited by 6
  13. 2023Dynamic nsNet2: Efficient Deep Noise Suppression with Early Exitingcited by 6
  14. 2022T-RECX: Tiny-Resource Efficient Convolutional neural networks with early-eXitcited by 5
  15. 2025A Survey on Dynamic Neural Networks: from Computer Vision to Multi-modal Sensor Fusioncited by 5
  16. 2022Understanding the Robustness of Multi-Exit Models under Common Corruptionscited by 4
  17. 2025Multi-Exit Kolmogorov-Arnold Networks: enhancing accuracy and parsimonycited by 4
  18. 2025Adaptive Slimming for Scalable and Efficient Speech Enhancementcited by 4
  19. 2022A Closer Look at Branch Classifiers of Multi-exit Architecturescited by 3
  20. 2023Anticipate, Ensemble and Prune: Improving Convolutional Neural Networks via Aggregated Early Exitscited by 3
  21. 2023FIANCEE: Faster Inference of Adversarial Networks via Conditional Early Exitscited by 3
  22. 2023Learning Task-preferred Inference Routes for Gradient De-conflict in Multi-output DNNscited by 3
  23. 2023Training dynamic models using early exits for automatic speech recognition on resource-constrained devicescited by 3
  24. 2023Adaptive Early Exiting for Collaborative Inference over Noisy Wireless Channelscited by 3
  25. 2024On the Impact of Black-box Deployment Strategies for Edge AI on Latency and Model Performancecited by 3
  26. 2024LightFF: Lightweight Inference for Forward-Forward Algorithmcited by 3
  27. 2025Beyond Greedy Exits: Improved Early Exit Decisions for Risk Control and Reliabilitycited by 3
  28. 2023Optimizing Attention and Cognitive Control Costs Using Temporally-Layered Architecturescited by 2
  29. 2023REDS: Resource-Efficient Deep Subnetworks for Dynamic Resource Constraintscited by 2
  30. 2024Decoding at the Speed of Thought: Harnessing Parallel Decoding of Lexical Units for LLMscited by 2
  31. 2025Early-Exit Graph Neural Networkscited by 2
  32. 2025Knowing When to Quit: Probabilistic Early Exits for Speech Separationcited by 2
  33. 2021High Frequency EEG Artifact Detection with Uncertainty via Early Exit Paradigmcited by 1
  34. 2022Binary Early-Exit Network for Adaptive Inference on Low-Resource Devicescited by 1
  35. 2022QuickNets: Saving Training and Preventing Overconfidence in Early-Exit Neural Architecturescited by 1
  36. 2022Temporally Layered Architecture for Adaptive, Distributed and Continuous Controlcited by 1
  37. 2023Dynamic Neural Network is All You Need: Understanding the Robustness of Dynamic Mechanisms in Neural Networkscited by 1
  38. 2023Using Early Exits for Fast Inference in Automatic Modulation Classificationcited by 1

50 most cited of 70 All 70 (JSON)