---
title: "Robust CurveMoE：基于模式连通性的混合专家模型多范数对抗防御"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.26043v1"
permalink: "https://feed.astralor.com/items/itm_e89b9e610fa0bbed58388863"
published: "2026-08-26T17:21:33.000Z"
score: 73
channel: "ai"
tags: ["Robust CurveMoE","对抗防御","模式连通性","混合专家"]
---

# Robust CurveMoE：基于模式连通性的混合专家模型多范数对抗防御

## AI 点评

> 这篇论文最值得关注的地方在于把模式连通性引入 MoE，用曲线路径自然解决了多范数鲁棒性冲突，思路非常聪明。对于研究对抗鲁棒性或者高效模型设计的人，这套选择性专家化和部分更新策略很有参考价值；对工程团队来说，如果复现成本可控，可能是比多模型集成更轻量的防御方案。

## 速览

多范数对抗防御旨在同时抵御不同范数约束下的对抗扰动，但现有方法常需在单一参数配置中优化互斥的鲁棒目标，训练成本高且存在权衡。Robust CurveMoE 提出一种高效混合专家框架，通过低损耗路径连接针对不同扰动范数特化的模型，并利用路径上模型的互补鲁棒性。该方法从鲁棒约束曲线位置提取干净与范数特化专家，仅对关键层进行专家化，共享其余参数；同时引入贡献引导的部分更新，基于初始化梯度分数选择关键曲线参数，并在理论上界定了部分与完全优化的目标差距。在 CIFAR-100 和 ImageNet-100 上以 WideResNet 和 Vision Transformer 进行实验，Robust CurveMoE 在干净准确率、范数特定准确率和联合准确率上均一致优于 MSD 与 ERMC，联合准确率较最强基线分别提升 2.37 和 2.13 个百分点。

## 关键点

- 提出 Robust CurveMoE 框架，利用低损耗路径连通不同范数特化模型，实现互补鲁棒性。
- 引入选择性专家化与贡献引导的部分更新，大幅降低曲线构造与训练成本。
- 在 CIFAR-100 和 ImageNet-100 上，Union 准确率较最强基线分别提升 2.37 与 2.13 个百分点。
- 提供部分更新与完全优化之间目标差距的理论界。

## 全文

Multi-norm adversarial defense aims to protect neural networks against perturbations defined by different norm constraints, but existing methods typically optimize competing robustness objectives within a single parameter configuration, leading to substantial training cost and unfavorable robustness trade-offs. We propose Robust CurveMoE, an efficient mixture-of-experts framework that connects models specialized for different perturbation norms through a low-loss path and exploits the complementary robustness profiles of models along this path. Robust CurveMoE derives clean and norm-specialized experts from robustness-constrained curve locations and selectively expertizes only influential layers, while sharing the remaining parameters across routing paths. To further reduce curve-construction cost, we introduce contribution-guided partial updating, which selects influential curve parameters using initialization-based gradient scores. We also theoretically bound the objective gap between partial and full curve optimization. Experiments on CIFAR-100 and ImageNet-100 with WideResNet and Vision Transformer architectures show that Robust CurveMoE consistently improves clean, norm-specific, and Union accuracy over MSD and ERMC. In particular, it improves Union accuracy by 2.37 and 2.13 percentage points over the strongest baseline on CIFAR-100 and ImageNet-100, respectively. Extensive ablations further validate the effectiveness of partial updating, selective expertization, and robustness-constrained expert selection.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_e89b9e610fa0bbed58388863