Saved in:
Bibliographic Details
Main Authors: Yudin, Nikolay, Gaponov, Alexander, Kudriashov, Sergei, Rakhuba, Maxim
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2507.07814
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916837256593408
author Yudin, Nikolay
Gaponov, Alexander
Kudriashov, Sergei
Rakhuba, Maxim
author_facet Yudin, Nikolay
Gaponov, Alexander
Kudriashov, Sergei
Rakhuba, Maxim
contents We present a novel local Lipschitz bound for self-attention blocks of transformers. This bound is based on a refined closed-form expression for the spectral norm of the softmax function. The resulting bound is not only more accurate than in the prior art, but also unveils the dependence of the Lipschitz constant on attention score maps. Based on the new findings, we suggest an explanation of the way distributions inside the attention map affect the robustness from the Lipschitz constant perspective. We also introduce a new lightweight regularization term called JaSMin (Jacobian Softmax norm Minimization), which boosts the transformer's robustness and decreases local Lipschitz constants of the whole network.
format Preprint
id arxiv_https___arxiv_org_abs_2507_07814
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Pay Attention to Attention Distribution: A New Local Lipschitz Bound for Transformers
Yudin, Nikolay
Gaponov, Alexander
Kudriashov, Sergei
Rakhuba, Maxim
Machine Learning
Numerical Analysis
15A42, 15A60, 68T07
We present a novel local Lipschitz bound for self-attention blocks of transformers. This bound is based on a refined closed-form expression for the spectral norm of the softmax function. The resulting bound is not only more accurate than in the prior art, but also unveils the dependence of the Lipschitz constant on attention score maps. Based on the new findings, we suggest an explanation of the way distributions inside the attention map affect the robustness from the Lipschitz constant perspective. We also introduce a new lightweight regularization term called JaSMin (Jacobian Softmax norm Minimization), which boosts the transformer's robustness and decreases local Lipschitz constants of the whole network.
title Pay Attention to Attention Distribution: A New Local Lipschitz Bound for Transformers
topic Machine Learning
Numerical Analysis
15A42, 15A60, 68T07
url https://arxiv.org/abs/2507.07814