mlsanityData to 14 Oct 2026

BlockDrop: Dynamic Inference Paths in Residual Networks

arXiv 1711.08393 ↗First posted 22 Nov 2017cs.CV cs.LGcited by 189

Not on the map. Why?

Abstract (arXiv)

Very deep convolutional neural networks offer excellent recognition results, yet their computational expense limits their impact for many real-world applications. We introduce BlockDrop, an approach that learns to dynamically choose which layers of a deep network to execute during inference so as to best reduce total computation without degrading prediction accuracy. Exploiting the robustness of Residual Networks (ResNets) to layer dropping, our framework selects on-the-fly which residual blocks to evaluate for a given novel image. In particular, given a pretrained ResNet, we train a policy network in an associative reinforcement learning setting for the dual reward of utilizing a minimal number of blocks while preserving recognition accuracy. We conduct extensive experiments on CIFAR and ImageNet. The results provide strong quantitative and qualitative evidence that these learned policies not only accelerate inference but also encode meaningful visual information. Built upon a ResNet-101 model, our method achieves a speedup of 20% on average, going as high as 36% for some images, while maintaining the same 76.4% top-1 accuracy on ImageNet.

Key facts. Cited by 189 papers in the map (data through 14 Oct 2026). First posted on arXiv 22 Nov 2017.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2018: 1; 2019: 0; 2020: 0; 2021: 48; 2022: 39; 2023: 37; 2024: 20; 2025: 16; 2026: 28.

189 papers.

Sort:

By area:

  1. 2023Ego-Exo4D: Understanding Skilled Human Activity from First- and Third-Person Perspectivescited by 490
  2. 2021Dynamic Neural Networks: A Surveycited by 408
  3. 2021AdaViT: Adaptive Tokens for Efficient Vision Transformercited by 362
  4. 2021Pruning and Quantization for Deep Neural Network Acceleration: A Surveycited by 263
  5. 2021AdaViT: Adaptive Vision Transformers for Efficient Image Recognitioncited by 228
  6. 2022Are Multimodal Transformers Robust to Missing Modality?cited by 186
  7. 2021Not All Images are Worth 16x16 Words: Dynamic Transformers for Efficient Image Recognitioncited by 144
  8. 2021IA-RED2: Interpretability-Aware Redundancy Reduction for Vision Transformerscited by 127
  9. 2021Learning Task-Oriented Communication for Edge Inference: An Information Bottleneck Approachcited by 110
  10. 2021MOOD: Multi-level Out-of-distribution Detectioncited by 94
  11. 2021Spatially-Adaptive Image Restoration using Distortion-Guided Networkscited by 77
  12. 2021Adaptive Inference through Early-Exit Networks: Design, Challenges and Directionscited by 75
38 more
  1. 2021FrameExit: Conditional Early Exiting for Efficient Video Recognitioncited by 59
  2. 2021A Survey on Green Deep Learningcited by 52
  3. 2021AdaFuse: Adaptive Temporal Fusion Network for Efficient Action Recognitioncited by 48
  4. 2024AIM: Adaptive Inference of Multi-Modal LLMs via Token Merging and Pruningcited by 43
  5. 2022Glance and Focus Networks for Dynamic Visual Recognitioncited by 41
  6. 2023SEENN: Towards Temporal Spiking Early-Exit Neural Networkscited by 41
  7. 2021Edge-Cloud Polarization and Collaboration: A Comprehensive Survey for AIcited by 39
  8. 2021AdaMML: Adaptive Multi-Modal Learning for Efficient Video Recognitioncited by 36
  9. 2021Dynamic Network Quantization for Efficient Video Inferencecited by 36
  10. 2022Efficient Split-Mix Federated Learning for On-Demand and In-Situ Customizationcited by 35
  11. 2024SpaceByte: Towards Deleting Tokenization from Large Language Modelingcited by 34
  12. 2022Enable Deep Learning on Mobile Devices: Methods, Systems, and Applicationscited by 32
  13. 2021Dynamic Resolution Networkcited by 31
  14. 2021Dynamic Transfer for Multi-Source Domain Adaptationcited by 29
  15. 2021MicroNet: Improving Image Recognition with Extremely Low FLOPscited by 26
  16. 2021Deep Joint Source-Channel Coding for Wireless Image Transmission with Adaptive Rate Controlcited by 26
  17. 2021Task2Sim : Towards Effective Pre-training and Transfer from Synthetic Datacited by 26
  18. 2023Latency-aware Unified Dynamic Networks for Efficient Image Recognitioncited by 26
  19. 2022DBQ-SSD: Dynamic Ball Query for Efficient 3D Object Detectioncited by 25
  20. 2023Green Edge AI: A Contemporary Surveycited by 25
  21. 2021Fire Together Wire Together: A Dynamic Pruning Approach with Self-Supervised Mask Predictioncited by 24
  22. 2023Dynamic Grained Encoder for Vision Transformerscited by 24
  23. 2024Prompt-Driven Dynamic Object-Centric Learning for Single Domain Generalizationcited by 24
  24. 2021On the Effect of Pruning on Adversarial Robustnesscited by 23
  25. 2025VORTA: Efficient Video Diffusion via Routing Sparse Attentioncited by 23
  26. 2024Layer- and Timestep-Adaptive Differentiable Token Compression Ratios for Efficient Diffusion Transformerscited by 22
  27. 2021LegoDNN: Block-grained Scaling of Deep Neural Networks for Mobile Visioncited by 20
  28. 2023Training a Large Video Model on a Single Machine in a Daycited by 20
  29. 2025Block-Recurrent Dynamics in Vision Transformerscited by 20
  30. 2023AdaptiveNet: Post-deployment Neural Architecture Adaptation for Diverse Edge Environmentscited by 19
  31. 2021Hard-Attention for Scalable Image Classificationcited by 18
  32. 2021Knowledge Evolution in Neural Networkscited by 18
  33. 2021Anytime Dense Prediction with Confidence Adaptivitycited by 18
  34. 2021Multi-Exit Semantic Segmentation Networkscited by 18
  35. 2022Efficient Deep Visual and Inertial Odometry with Adaptive Visual Modality Selectioncited by 18
  36. 2023GPFL: Simultaneously Learning Global and Personalized Feature Information for Personalized Federated Learningcited by 18
  37. 2022Dynamic Spatio-Temporal Specialization Learning for Fine-Grained Action Recognitioncited by 17
  38. 2022DeepPerform: An Efficient Approach for Performance Testing of Resource-Constrained Neural Networkscited by 17

50 most cited of 189 All 189 (JSON)