Literature Notes
/
Multi-class N-pair loss (Sohn 2016) is a generalization of triplet loss* allowing joint comparision‣ among more than one negative samples. When applied on a pair of positive samples 𝐳_𝐢 and 𝐳_𝐣 (𝑦_𝑖=𝑦_𝑗), where z_i z_j is the inner product, which is equivalent to cosine similarity when both vectors have unit norm.
/
그림 설명
Search
Share
그림 설명
Default view
Search
input
type
neural network 의 output
위 그림에서
v1
Open
anchor
f(v1)
f
v2
Open
negative
f(v2)
f+
v3
Open
positive
f(v3)
f-
Open
COUNT
4