Unverified50% confidenceFactExact time
批归一化由Ioffe和Szegedy于2015年提出，核心机制是对小批量数据的均值和方差进行归一化并学习缩放参数γ和平移参数β
1
Sources
50%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
疟疾细胞AI识别：医学影像分类实践全解析
Related Claims
UnverifiedBatch Normalization was proposed by Ioffe and Szegedy in 2015, standardizing activation values within a mini-batch.65% similarUnverified可说明性（Explainability）催生了LIME、SHAP等事后归因方法，面向外部用户回答为什么模型给出该输出64% similarUnverified批归一化于2015年提出，通过在每层对激活值进行标准化来稳定梯度的分布64% similarVerified混合专家架构MoE通过稀疏激活机制，推理时只激活模型总参数的1/8至1/463% similarUnverifiedThe model ensemble effect of integrating answers from multiple models can smooth out the systematic bias of a single model.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/507550API
curl https://kongchang.com/api/v1/knowledge/claims/507550MCP
get_claim(id=507550)