[KongchangAI]
Conceptnegative log-odds / 负对数几率损失

Negative Log-Odds

A loss form in classification tasks that is sensitive to the margin between the correct class and the strongest incorrect class. Its gradient concentrates on tokens that actually change the prediction tendency, and it is commonly used for gradient computation in data attribution.

Timeline (last 90 days)

Oct 30

Negative log-odds is defined as the negative of the difference between the log probability of the correct class and the log probability of the incorrect class.

Unverified50%

All Facts (1)

Source Articles