Socrates Loss unifies classification and confidence calibration in neural networks by adding an auxiliary unknown class and a dynamic uncertainty penalty to the loss function.
The loss at epoch 31 is⇒At epoch30, the classifier outputs[0.9,0.05,0.05]confidences, resulting int i,yi,e−1= 0.9based on previous predictions
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Socrates Loss: Unifying Confidence Calibration and Classification by Leveraging the Unknown
Socrates Loss unifies classification and confidence calibration in neural networks by adding an auxiliary unknown class and a dynamic uncertainty penalty to the loss function.