AI Glossary Term
What does “Training Loss” mean?
Definition
The model error value computed during training and optimized downward over time.
How Training Loss works
Example
During training, a dashboard plots the loss on the data the model is learning from. A steadily falling curve suggests learning, while a flat or spiking one signals trouble.
Why it matters
Training loss shows whether training is working, but by itself it cannot show how well the model will perform on new data.
Common questions
Training loss vs validation loss?
Training loss is measured on the data used to learn. Validation loss is measured on held-out data and shows generalization.
What if it stops dropping?
The model may have converged, the learning rate may be off, or capacity may be limited.
Should it reach zero?
Rarely. Reaching zero on real data usually signals overfitting.