MoE模型虽冗余但脆弱，输出稀释导致道德信息易受干扰。
Output Dilution: Redundant but Fragile Representations in MoE Models
- MoE通过多专家平均输出，造成信号大幅衰减
- 道德信息在噪声下准确率下降4.2倍，远逊于密集模型
- 架构缺陷致冗余不等于鲁棒，适合关注模型稳定性的研究者
Mixture-of-Experts (MoE) 模型看似与密集模型一样稳健地编码道德内容，实则更易受扰动。在 OLMoE-1B-7B 中，线性探测器可从几乎任意专家-层组合中恢复道德极性，平均峰值层准确率超过90%。然而，在激活噪声下的鲁棒性相差4.2倍，密集模型能承受的噪声水平，MoE模型却难以应对。我们将其归因于输出稀释：由于 MoE 块在贡献残差流前对活跃专家求平均，传递至下游层的前馈信号比同规模密集MLP小近两个数量级。道德信息虽在聚合中保留，但其规模微弱，极易被扰动淹没。路由机制本身在噪声下仍稳定，脆弱性完全源于稀释后的聚合结果。检查点轨迹证实此为架构本质问题，非训练所致：专家从未专业化，准确率在数千步内即饱和。
原文摘要 · Abstract (English)
Mixture-of-Experts (MoE) models appear to encode moral content as robustly as dense models, yet prove far more fragile in their encoding. In OLMoE-1B-7B, linear probes recover moral valence from nearly every expert-layer combination, with mean peak-layer accuracy above 90%. But these representations collapse under levels of activation noise that a dense model of matched size easily tolerates, with a 4.2-fold difference in robustness. We trace this to output dilution. Because the MoE block averages across active experts before contributing to the residual stream, the feedforward signal reaching downstream layers is nearly two orders of magnitude smaller than in a dense MLP. Moral information, our interest, survives aggregation intact but at a scale trivially overwhelmed by perturbation. Routing itself remains stable under noise while the vulnerability originates entirely in the diluted aggregate. Checkpoint trajectories confirm this is architectural, not learned. Experts never specialize and accuracy saturates within the first few thousand steps. In sparse architectures, redundant encoding does not imply robust encoding.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。