Unverified50% confidenceFactExact time
对已做L2归一化的嵌入，余弦距离与欧氏距离给出的排序结果完全一致
1
Sources
50%
Confidence
Long-term
Relevance
7/10/2026
First Seen
Sources
DINOv2 vs SigLIP：k-NN分类为何差距悬殊？视觉编码器选型避坑指南
redditr/MachineLearning7/8/2026
Related Claims
Unverified当特征向量被L2归一化后，余弦相似度等价于点积运算72% similarUnverified轨迹相似度可采用基于编辑距离（Levenshtein Distance）的动作序列比对或基于余弦相似度的向量化比对64% similarVerified从贝叶斯视角理解，L1正则化对应参数的拉普拉斯先验，L2正则化对应高斯先验61% similarUnverified两个不可约表示的张量积l₁⊗l₂可分解为|l₁-l₂|到l₁+l₂之间所有整数阶不可约表示的直和，CG系数给出具体的基变换矩阵59% similarUnverified在满足约束等距性（RIP）的字典下，L1松弛（LASSO）可以精确恢复稀疏解，为L1正则化提供理论保证58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/456350API
curl https://kongchang.com/api/v1/knowledge/claims/456350MCP
get_claim(id=456350)