Verified65% confidenceFactExact time
MoE (Mixture of Experts) is a neural network architecture that splits a large feedforward network into multiple parallel expert subnetworks, using a router to dynamically select and activate a few experts based on the input, thereby decoupling total parameter count from activated parameters.
3
Sources
65%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
Unverified混合专家（MoE）架构最早由 Jacobs 等人于 1991 年提出，其核心思想是将复杂模型拆分为多个专家子网络，由门控网络根据输入动态决定各专家权重组合71% similarUnverifiedThe Mixture of Experts (MoE) model is a classic machine learning concept originating in 1991.65% similarUnverifiedThe core idea of the Mixture of Experts (MoE) architecture originates from the divide-and-conquer learning theory of 1991.62% similarUnverifiedMoE思想源自1991年统计学习领域，经Google Brain团队在2017年Sparsely-Gated MoE论文中工程化改造后进入深度学习实践62% similarUnverified任务路由的思想根源可追溯到早期专家系统和集成学习中的模型选择与混合专家（MoE）机制61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1032743API
curl https://kongchang.com/api/v1/knowledge/claims/1032743MCP
get_claim(id=1032743)