Cost Function

In short

Tells us how close the Model’s prediction is to the expected value. The closer, the better.

Like a teacher grading your answer — it tells you how far off you were from the correct one.

During Training, a Model learns by making predictions and then checking how wrong it was. The cost function is what does that checking — after the model makes a prediction, it measures how well or how badly it did. This score is then used by Gradient Descent to figure out how to update the Model Parameters to improve next time. The whole Training loop revolves around minimizing this cost.

Related

• Training - cost function drives the training loop
• Gradient Descent - uses the cost to update parameters
• Model Parameters - get adjusted based on cost