Unverified50% confidenceFactExact time
Greg Yang等人提出的μP（Maximal Update Parameterization）理论试图在保留特征学习的同时实现理论分析
1
Sources
50%
Confidence
Long-term
Relevance
8/22/2026
First Seen
Sources
Related Claims
Unverified集成学习的理论基础来自偏差-方差权衡框架，单一复杂模型存在高方差而简单模型存在高偏差66% similarUnverifiedHimanshu指出LLM推理管线的最大优势在于可微分，可用强化学习优化，但无法学习代数构造65% similarUnverifiedFrontier models commonly adopt a Mixture of Experts (MoE) architecture that activates only a subset of parameters during inference to balance speed and quality64% similarUnverified凸优化理论中的KKT条件是理解支持向量机SVM对偶问题以及RLHF中约束优化设计的必要工具64% similarUnverifiedYang Song et al. unified the diffusion process under the framework of continuous-time stochastic differential equations and proposed score-based generative models.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/785593API
curl https://kongchang.com/api/v1/knowledge/claims/785593MCP
get_claim(id=785593)