Lipschitz verification of neural networks through training

View PDF HTML (experimental)

Abstract:The global Lipschitz constant of a neural network governs both adversarial robustness and generalization. Conventional approaches to

certified training" typically follow a train-then-verify paradigm: they train a network and then attempt to bound its Lipschitz constant. Because the efficient 

certified training" typically follow a train-then-verify paradigm: they train a network and then attempt to bound its Lipschitz constant. Because the efficient 

trivial bound" (the product of the layerwise Lipschitz constants) is exponentially loose for arbitrary networks, these approaches must rely on computationally expensive techniques such as semidefinite programming, mixed-integer programming, or branch-and-bound. We propose a different paradigm: rather than designing complex verifiers for arbitrary networks, we design networks to be verifiable by the fast trivial bound. We show that directly penalizing the trivial bound during training forces it to become tight, thereby effectively regularizing the true Lipschitz constant. To achieve this, we identify three structural obstructions to a tight trivial bound (dead neurons, bias terms, and ill-conditioned weights) and introduce architectural mitigations, including a novel notion of norm-saturating polyactivations and bias-free sinusoidal layers. Our approach avoids the runtime complexity of advanced verification while achieving strong results: we train robust networks on MNIST with Lipschitz bounds that are small (orders of magnitude lower than comparable works) and tight (within 10% of the ground truth). The experimental results validate the theoretical guarantees, support the proposed mechanisms, and extend empirically to diverse activations and non-Euclidean norms.

Subjects:

Machine Learning (cs.LG)

Cite as: arXiv:2603.28113 [cs.LG]

(or arXiv:2603.28113v1 [cs.LG] for this version)

https://doi.org/10.48550/arXiv.2603.28113

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Simon Kuang [view email] [v1] Mon, 30 Mar 2026 07:20:50 UTC (1,849 KB)