Unverified50% confidenceFactExact time
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
1
Sources
50%
Confidence
Long-term
Relevance
8/21/2026
First Seen
Sources
Related Claims
UnverifiedBatch Normalization was proposed by Ioffe and Szegedy in 2015, standardizing activation values within a mini-batch.66% similarUnverified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β64% similarUnverified优先固定特征与归一化，再逐步拆解架构组件，才能得到真正可信、可复现的量化研究结论63% similarUnverified梯度提升是当前结构化数据性能最强的系列方法，通过迭代拟合残差逐步提升预测精度62% similarUnverifiedClassification predicts discrete class labels, while regression predicts continuous numerical values.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/782763API
curl https://kongchang.com/api/v1/knowledge/claims/782763MCP
get_claim(id=782763)