AI engineering glossary · Evaluation metrics
What is temperature scaling?
Temperature scaling: A calibration method for neural networks that divides the model's logits by a single learned number T before the softmax. T above 1 softens overconfident probabilities and T below 1 sharpens them. It leaves the predicted class unchanged, so accuracy stays the same.
Temperature scaling: an example
A classifier's 99% confidence drops to 85% after dividing its logits by T = 2, better matching its real accuracy.
Related terms
More Evaluation metrics terms
- Token-level F1
- Translation edit rate
- True negative
- True positive
- V-measure
- Weighted averaging
- Weighted kappa
- Word error rate
Where you learn it
Temperature scaling is part of the AI Engineering Program, which teaches it with notes, a hands-on lab and an exam. The first lesson is free.