Unverified50% confidenceSolutionExact time
研究者引入辅助负载均衡损失来惩罚专家负载不均匀分布，DeepSeek系列还引入专家亲和度机制
1
Sources
50%
Confidence
Long-term
Relevance
7/13/2026
First Seen
Sources
15岁少年从零训练200M MoE语言模型：完整架构与实战拆解
redditr/learnmachinelearning7/12/2026
Related Claims
UnverifiedThe MoE architecture faces load balancing challenges, and researchers typically introduce auxiliary loss functions to mitigate the expert collapse problem, while teams such as DeepSeek have explored auxiliary-loss-free balancing strategies.73% similarUnverifiedAs an early research contribution from the technical community, the dispersion loss still requires further validation in areas such as hyperparameter settings, cross-task generalizability, and comparison with existing methods.69% similarUnverified梯度下降的收敛条件与学习率和损失函数的Lipschitz常数的关系有关62% similarUnverified安全研究者将系统行动能力远超其安全判断能力的潜在风险称为能力越界（Capability Overhang）61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/502527API
curl https://kongchang.com/api/v1/knowledge/claims/502527MCP
get_claim(id=502527)