arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.15091stat.MLcs.ITcs.LGmath.IT
收藏

专家混合架构在有限速率门控下的应用:通信与泛化之间的权衡

Mixture-of-Experts under Finite-Rate Gating: Communication--Generalization Trade-offs

Institut Polytechnique des Sciences Avancées (IPSA) and LINCS Lab(先进科学研究院(IPSA)和LINCS实验室)

Ali Khalesi, Mohammad Reza Deylam Salehi

AI总结:

本文从信息论角度分析了专家混合架构的门控机制,推导了有限速率门控的率-失真特性,并通过实验验证了门控速率、表达能力和泛化能力之间的权衡。

摘要:

专家混合(MoE)架构将预测任务分解为由门控机制选择的专用专家子网络。本文采用通信理论观点,将门控建模为在有限信息速率下工作的随机信道。在信息论学习框架下,专门化了一个互信息泛化界限,并开发了有限速率门控的率-失真特性D(R_g),其中R_g:=I(X; T),从而在标准经验率-失真最优条件下得到E[R(W)] ≤ D(R_g)+δ_m+√((2/m) I(S; W))。分析结果给出了通信受限的MoE系统的容量感知限制,并通过合成多专家模型的数值模拟验证了门控速率、表达能力和泛化能力之间的预测权衡。

英文摘要:

Mixture-of-Experts (MoE) architectures decompose prediction tasks into specialized expert sub-networks selected by a gating mechanism. This letter adopts a communication-theoretic view of MoE gating, modeling the gate as a stochastic channel operating under a finite information rate. Within an information-theoretic learning framework, {we specialize a mutual-information generalization bound and develop a rate-distortion characterization $D(R_g)$ of finite-rate gating, where $R_g:=I(X; T)$, yielding (under a standard empirical rate-distortion optimality condition) $\mathbb{E}[R(W)] \le D(R_g)+δ_m+\sqrt{(2/m)\, I(S; W)}$. }The analysis yields capacity-aware limits for communication-constrained MoE systems, and numerical simulations on synthetic multi-expert models empirically confirm the predicted trade-offs between gating rate, expressivity, and generalization.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑