---
title: "Robust CurveMoE: Multi-Norm Adversarial Defense for Mixture-of-Experts Models via Mode Connectivity"
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","Adversarial Defense","Mode Connectivity","Mixture-of-Experts"]
---

# Robust CurveMoE: Multi-Norm Adversarial Defense for Mixture-of-Experts Models via Mode Connectivity

## AI commentary

> The most interesting aspect of this paper is introducing mode connectivity into MoE, naturally resolving multi-norm robustness conflicts via a curved path—quite a clever idea. For researchers in adversarial robustness or efficient model design, the selective expertization and partial updating strategies offer valuable reference; for engineering teams, if reproduction cost is manageable, this could be a lighter-weight defense alternative to multi-model ensembles.

## At a glance

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.

## Key points

- Proposes Robust CurveMoE, a framework that connects norm-specialized models via a low-loss path for complementary robustness.
- Introduces selective expertization and contribution-guided partial updating to significantly reduce curve construction and training costs.
- Improves Union accuracy over the strongest baseline by 2.37 and 2.13 percentage points on CIFAR-100 and ImageNet-100, respectively.
- Provides a theoretical bound on the objective gap between partial and full curve optimization.

## Full text

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.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_e89b9e610fa0bbed58388863