[KongchangAI]
ConceptSymmetric Cross Entropy / SCE

Symmetric Cross-Entropy

Symmetric cross-entropy is a robust loss function with theoretical tolerance to label noise, formed by combining cross-entropy with reverse cross-entropy, commonly used for classification training under noisy labels.

Timeline (last 90 days)

Nov 7

Loss functions such as Symmetric Cross Entropy (SCE) have theoretical tolerance for symmetric noise, but their effect on asymmetric noise is limited.

Unverified50%
Nov 7

Symmetric Cross Entropy (SCE) is composed of cross entropy and reverse cross entropy (RCE).

Unverified50%

All Facts (2)

Source Articles