Efficient Bound of Lipschitz Constant for Convolutional Layers by Gram Iteration

Blaise Delattre, Quentin Barthélemy, Alexandre Araujo, Alexandre Allauzen
5/25/2023
cs.LGcs.AI

Official code: blaisedelattre/lip4conv (16 stars, MIT, PyTorch, last commit 2024-02-02).

Abstract

Since the control of the Lipschitz constant has a great impact on the training stability, generalization, and robustness of neural networks, the estimation of this value is nowadays a real scientific challenge. In this paper we introduce a precise, fast, and differentiable upper bound for the spectral norm of convolutional layers using circulant matrix theory and a new alternative to the Power iteration. Called the Gram iteration, our approach exhibits a superlinear convergence. First, we show through a comprehensive set of experiments that our approach outperforms other state-of-the-art methods in terms of precision, computational cost, and scalability. Then, it proves highly effective for the Lipschitz regularization of convolutional neural networks, with competitive results against concurrent approaches. Code is available at https://github.com/blaisedelattre/lip4conv.

View on arXivView PDF

Code Implementations(1)

blaisedelattre/lip4convOfficial100%
PyTorchMIT

Code for Spectral Norm of Convolutional Layers with Circular and Zero Paddings and Efficient Bound of Lipschitz Constant for Convolutional Layers by Gram Iteration

163Python, Jupyter NotebookMay 26, 20232 years agoMIT
convolutional-layerslipschitz-constantlipschitz-networklipschitz-regularizationnorm+2 more

Similar Papers

Spectral Norm of Convolutional Layers with Circular and Zero Paddings

89% match1 repo162024

Computable Lipschitz Bounds for Deep Neural Networks

85% match2024

Scalable Lipschitz Estimation for CNNs

85% match2024

ECLipsE: Efficient Compositional Lipschitz Constant Estimation for Deep Neural Networks

85% match1 repo32024

LipKernel: Lipschitz-Bounded Convolutional Neural Networks via Dissipative Layers

84% match1 repo22024

ECLipsE-Gen-Local: Efficient Compositional Local Lipschitz Estimates for Deep Neural Networks

83% match2025

Cite this paper

@article{delattre2023efficient,
  title  = {Efficient Bound of Lipschitz Constant for Convolutional Layers by Gram Iteration},
  author = {Blaise Delattre and Quentin Barthélemy and Alexandre Araujo and Alexandre Allauzen},
  year   = {2023},
  eprint = {2305.16173},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2305.16173v3},
  journal = {ICML 2023 2023}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2305.16173.svg)](https://codeofpaper.com/paper/2305.16173)

Discussion