#regularization-strength

Question 1: Single choice

You raise the L2 regularization strength $\lambda$ far above the value that gave the best validation loss. What do you expect to see?

Training and validation losses both increase, because the model underfits1
Training loss falls while validation loss rises, because the model overfits2
Both losses fall toward zero, because smaller weights always generalize better3
Only the validation loss changes; the training loss is unaffected by λ4