---
title: "Symmetry Breaking at the Edge of Stability via Central Flow Analysis"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2611.06718v1"
permalink: "https://feed.astralor.com/items/itm_2dd57eab6678ac6f4b0bb565"
published: "2026-11-04T16:11:59.000Z"
score: 56
channel: "tech"
tags: ["Symmetry Breaking","Edge","Stability","Central Flow Analysis","Continuous","ReLU","Finite-step"]
---

# Symmetry Breaking at the Edge of Stability via Central Flow Analysis

## 速览

Continuous symmetries in neural networks lead to conserved quantities that are invariant under gradient flow, such as layer balancedness in linear and ReLU networks. Finite-step gradient descent need not preserve these …

## 关键点

- Continuous symmetries in neural networks lead to conserved quantities that are invariant under gradient flow, such as layer balancedness in linear and ReLU networks.
- Finite-step gradient descent need not preserve these laws, particularly at the edge of stability (EOS), where the iterates oscillate around a smoother trajectory, which can be described by the central flow dynamics (Coh…
- We study how gradient flow conservation laws evolve under the central flow description of this regime.

## 全文

Continuous symmetries in neural networks lead to conserved quantities that are invariant under gradient flow, such as layer balancedness in linear and ReLU networks. Finite-step gradient descent need not preserve these laws, particularly at the edge of stability (EOS), where the iterates oscillate around a smoother trajectory, which can be described by the central flow dynamics (Cohen et al., 2025). We study how gradient flow conservation laws evolve under the central flow description of this regime. We show that the curvature correction induces a finite-dimensional drift determined by the oscillation covariance along critical parameter directions. We thus take the first steps to analyse which conservation laws remain and are broken at the EOS according to the central flow dynamics by analysing a scalar two-layer model, two-layer matrix factorisation, and a two-layer ReLU network, where the known conservation laws are shown to be complete (Marcotte et al., 2023).

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_2dd57eab6678ac6f4b0bb565