mlsanityData to 7 Oct 2026

Spectral Leakage and Rethinking the Kernel Size in CNNs

arXiv 2101.10143 ↗First posted 25 Jan 2021cs.CV cs.LGcited by 11

Not on the map. Why?

Abstract (arXiv)

Convolutional layers in CNNs implement linear filters which decompose the input into different frequency bands. However, most modern architectures neglect standard principles of filter design when optimizing their model choices regarding the size and shape of the convolutional kernel. In this work, we consider the well-known problem of spectral leakage caused by windowing artifacts in filtering operations in the context of CNNs. We show that the small size of CNN kernels make them susceptible to spectral leakage, which may induce performance-degrading artifacts. To address this issue, we propose the use of larger kernel sizes along with the Hamming window function to alleviate leakage in CNN architectures. We demonstrate improved classification accuracy on multiple benchmark datasets including Fashion-MNIST, CIFAR-10, CIFAR-100 and ImageNet with the simple use of a standard window function in convolutional layers. Finally, we show that CNNs employing the Hamming window display increased robustness against various adversarial attacks.

Key facts. Cited by 11 papers in the map (data through 7 Oct 2026). Builds on 35 of its 66 references that are in the map. First posted on arXiv 25 Jan 2021.

Builds on

35 of 66 references.

References · 35

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2014Adam: A Method for Stochastic Optimizationcited by 36,960
  3. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  4. 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,267
  5. 2014ImageNet Large Scale Visual Recognition Challengecited by 8,441
  6. 2014Explaining and Harnessing Adversarial Examplescited by 7,287
  7. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 6,979
  8. 2015Rethinking the Inception Architecture for Computer Visioncited by 5,451
  9. 2013Intriguing properties of neural networkscited by 5,114
  10. 2017Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithmscited by 4,283
  11. 2014Going Deeper with Convolutionscited by 4,231
  12. 2015Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Codingcited by 2,291
23 more
  1. 2013Visualizing and Understanding Convolutional Networkscited by 2,278
  2. 2015DeepFool: a simple and accurate method to fool deep neural networkscited by 1,315
  3. 2019Adversarial Examples Are Not Bugs, They Are Featurescited by 950
  4. 2016Practical Black-Box Attacks against Machine Learningcited by 893
  5. 2018Robustness May Be at Odds with Accuracycited by 873
  6. 2017Evasion Attacks against Machine Learning at Test Timecited by 686
  7. 2017One pixel attack for fooling deep neural networkscited by 579
  8. 2016Delving into Transferable Adversarial Examples and Black-box Attackscited by 530
  9. 2019Making Convolutional Networks Shift-Invariant Againcited by 390
  10. 2019On the Relationship between Self-Attention and Convolutional Layerscited by 299
  11. 2016Harmonic Networks: Deep Translation and Rotation Equivariancecited by 293
  12. 2018Adversarial Robustness Toolbox v1.0.0cited by 214
  13. 2020On Translation Invariance in CNNs: Convolutional Layers can Exploit Absolute Spatial Locationcited by 141
  14. 2019Orthogonal Convolutional Neural Networkscited by 122
  15. 2013Fast Training of Convolutional Networks through FFTscited by 89
  16. 2019Superposition of many models into onecited by 64
  17. 2017Gabor Convolutional Networkscited by 55
  18. 2014Deep Roto-Translation Scattering for Object Classificationcited by 49
  19. 2020Towards Understanding Hierarchical Learning: Benefits of Neural Representationscited by 46
  20. 2018Scattering Networks for Hybrid Representation Learningcited by 33
  21. 2016Structured Receptive Fields in CNNscited by 32
  22. 2019Blurring the Line Between Structure and Learning to Optimize and Adapt Receptive Fieldscited by 11
  23. 2020Hierarchically Compositional Tasks and Deep Convolutional Networkscited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 2; 2022: 1; 2023: 4; 2024: 0; 2025: 3; 2026: 1.

11 papers.

  1. 2021Zero-Shot Day-Night Domain Adaptation with a Physics Priorcited by 47
  2. 2023As large as it gets: Learning infinitely large Filters via Neural Implicit Functions in the Fourier Domaincited by 8
  3. 2021Frequency learning for structured CNN filters with Gaussian fractional derivativescited by 6
  4. 2023SMPConv: Self-moving Point Representations for Continuous Convolutioncited by 6
  5. 2022PSO-Convolutional Neural Networks with Heterogeneous Learning Ratecited by 2
  6. 2023On the Interplay of Convolutional Padding and Adversarial Robustnesscited by 1
  7. 2026Analysis of Long Range Dependency Understanding in State Space Modelscited by 1
  8. 2023Fix your downsampling ASAP! Be natively more robust via Aliasing and Spectral Artifact free Poolingcited by 0
  9. 2025ISAC: An Invertible and Stable Auditory Filter Bank with Customizable Kernels for ML Integrationcited by 0
  10. 2025Finding Optimal Kernel Size and Dimension in Convolutional Neural Networks An Architecture Optimization Approachcited by 0
  11. 2025LAID: Lightweight AI-Generated Image Detection in Spatial and Spectral Domainscited by 0