mlsanityData to 7 Oct 2026

Do All MobileNets Quantize Poorly? Gaining Insights into the Effect of Quantization on Depthwise Separable Convolutional Networks Through the Eyes of Multi-scale Distributional Dynamics

arXiv 2104.11849 ↗First posted 24 Apr 2021cs.CV cs.LG eess.IVcited by 12

Not on the map. Why?

Abstract (arXiv)

As the "Mobile AI" revolution continues to grow, so does the need to understand the behaviour of edge-deployed deep neural networks. In particular, MobileNets are the go-to family of deep convolutional neural networks (CNN) for mobile. However, they often have significant accuracy degradation under post-training quantization. While studies have introduced quantization-aware training and other methods to tackle this challenge, there is limited understanding into why MobileNets (and potentially depthwise-separable CNNs (DWSCNN) in general) quantize so poorly compared to other CNN architectures. Motivated to gain deeper insights into this phenomenon, we take a different strategy and study the multi-scale distributional dynamics of MobileNet-V1, a set of smaller DWSCNNs, and regular CNNs. Specifically, we investigate the impact of quantization on the weight and activation distributional dynamics as information propagates from layer to layer, as well as overall changes in distributional dynamics at the network level. This fine-grained analysis revealed significant dynamic range fluctuations and a "distributional mismatch" between channelwise and layerwise distributions in DWSCNNs that lead to increasing quantized degradation and distributional shift during information propagation. Furthermore, analysis of the activation quantization errors show that there is greater quantization error accumulation in DWSCNN compared to regular CNNs. The hope is that such insights can lead to innovative strategies for reducing such distributional dynamics changes and improve post-training quantization for mobile.

Key facts. Cited by 12 papers in the map (data through 7 Oct 2026). Builds on 27 of its 35 references that are in the map. First posted on arXiv 24 Apr 2021.

Builds on

27 of 35 references.

References · 27

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 6,979
  3. 2018MobileNetV2: Inverted Residuals and Linear Bottleneckscited by 4,773
  4. 2017MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applicationscited by 3,827
  5. 2015Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classificationcited by 3,590
  6. 2016Wide Residual Networkscited by 3,057
  7. 2015Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Codingcited by 2,291
  8. 2019Searching for MobileNetV3cited by 1,984
  9. 2017Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inferencecited by 1,548
  10. 2017ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devicescited by 1,194
  11. 2016Pruning Filters for Efficient ConvNetscited by 1,129
  12. 2016SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model sizecited by 1,054
15 more
  1. 2020Up or Down? Adaptive Rounding for Post-Training Quantizationcited by 600
  2. 2017Learning Sparse Neural Networks through L0 Regularizationcited by 571
  3. 2018Quantizing deep convolutional networks for efficient inference: A whitepapercited by 507
  4. 2018FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural Architecture Searchcited by 480
  5. 2019Data-Free Quantization Through Weight Equalization and Bias Correctioncited by 328
  6. 2014Compressing Deep Convolutional Networks using Vector Quantizationcited by 189
  7. 2016Convolutional Neural Networks using Logarithmic Data Representationcited by 97
  8. 2019Trained Quantization Thresholds for Accurate and Efficient Fixed-Point Inference of Deep Neural Networkscited by 87
  9. 2020Robust Quantization: One Model to Rule Them Allcited by 63
  10. 2018SqueezeNext: Hardware-Aware Neural Network Designcited by 61
  11. 2019DeepHoyer: Learning Sparser Neural Network with Differentiable Scale-Invariant Sparsity Measurescited by 56
  12. 2019Same, Same But Different - Recovering Neural Network Quantization Error Through Weight Factorizationcited by 45
  13. 2018A Quantization-Friendly Separable Convolution for MobileNetscited by 22
  14. 2020Where Should We Begin? A Low-Level Exploration of Weight Initialization Impact on Quantized Behaviour of Deep Neural Networkscited by 2
  15. 2020FactorizeNet: Progressive Depth Factorization for Efficient Network Architecture Exploration Under Quantization Constraintscited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 1; 2022: 4; 2023: 4; 2024: 0; 2025: 2; 2026: 1.

12 papers.

  1. 2022Make RepVGG Greater Again: A Quantization-aware Approachcited by 21
  2. 2023Solving Oscillation Problem in Post-Training Quantization Through a Theoretical Perspectivecited by 17
  3. 2023Design Space Exploration of Low-Bit Quantized Neural Networks for Visual Place Recognitioncited by 10
  4. 2022Accelerating Deep Learning Model Inference on Arm CPUs with Ultra-Low Bit Quantization and Runtimecited by 6
  5. 2021An Underexplored Dilemma between Confidence and Calibration in Quantized Neural Networkscited by 2
  6. 2022Bit-serial Weight Pools: Compression and Arbitrary Precision Execution of Neural Networks on Resource Constrained Processorscited by 2
  7. 2022GHN-Q: Parameter Prediction for Unseen Quantized Convolutional Architectures via Graph Hypernetworkscited by 1
  8. 2023On Calibration of Modern Quantized Efficient Neural Networkscited by 1
  9. 2025LeanTTA: A Backpropagation-Free and Stateless Approach to Quantized Test-Time Adaptation on Edge Devicescited by 1
  10. 2025Starting Positions Matter: A Study on Better Weight Initialization for Neural Network Quantizationcited by 1
  11. 2026Beyond Layer Importance in Layer-wise Sparsity: An Inter-Layer Perturbation-Absorption Perspectivecited by 1
  12. 2023GHN-QAT: Training Graph Hypernetworks to Predict Quantization-Robust Parameters of Unseen Limited Precision Neural Networkscited by 0