arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2601.09043stat.MLcs.LG

马蹄混合专家(HS-MoE)模型

Horseshoe Mixtures-of-Experts (HS-MoE)

Booth School of Business University of Chicago(芝加哥大学商学院);Volgenau School of Engineering, George Mason University(乔治·马歇尔大学工程学院);Department of Systems Engineering and Operations Research George Mason University(乔治·马歇尔大学系统工程与运营管理系)

Nick Polson, Vadim Sokolov

AI总结:

HS-MoE模型通过结合马蹄先验和输入依赖门控,实现数据自适应的稀疏专家选择,并提出粒子学习算法用于顺序推断。

摘要:

马蹄混合专家(HS-MoE)模型提供了一种贝叶斯框架,用于在混合专家架构中实现稀疏专家选择。我们结合了马蹄先验的自适应全局-局部收缩与输入依赖的门控,从而实现数据自适应的专家使用稀疏性。我们的主要方法贡献是一种粒子学习算法,用于顺序推断,在此算法中,滤波器在时间上向前传播,同时仅跟踪充分统计量。我们还讨论了HS-MoE与现代混合专家层在大语言模型中的关系,这些层在部署时受到极端稀疏性约束(例如,每个token从大量专家池中激活少量专家)。

英文摘要:

Horseshoe mixtures-of-experts (HS-MoE) models provide a Bayesian framework for sparse expert selection in mixture-of-experts architectures. We combine the horseshoe prior's adaptive global-local shrinkage with input-dependent gating, yielding data-adaptive sparsity in expert usage. Our primary methodological contribution is a particle learning algorithm for sequential inference, in which the filter is propagated forward in time while tracking only sufficient statistics. We also discuss how HS-MoE relates to modern mixture-of-experts layers in large language models, which are deployed under extreme sparsity constraints (e.g., activating a small number of experts per token out of a large pool).

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。

↑