Case context: You are fitting a forecasting network on historical sales records. The training loss keeps getting smaller, but after several passes through the data, the validation loss starts to climb. To curb overfitting, you decide to end the optimization run once the validation metric stops improving.

Question: What method are you using, which optimization procedure is being cut short, and what bias–variance effect should you expect?

Sample answer: This is early stopping. You interrupt gradient descent before it fully converges by watching the validation loss. The usual consequence is lower variance and higher bias.

Key points:

• Method: early stopping
• Procedure cut short: gradient descent
• Effect on variance: it decreases
• Effect on bias: it increases

Rubric: The learner must identify early stopping, name gradient descent as the procedure being stopped, and state that variance decreases while bias increases.