DSA Studio

Loss Functions Checkpoint

MSE, cross-entropy, and loss intuition.

Get every question correct to pass this checkpoint. You can review feedback, revise your answers, and submit again.

1. Targets [1,3], predictions [2,1]. What is mean squared error averaged over the two examples?

Select one answer.

2. For a positive binary target, prediction probability changes from 0.5 to 0.9. What happens to cross-entropy?

Select one answer.

3. Model A has lower training loss than B but worse held-out loss. What does training loss alone establish?

Select one answer.

4. For y=0 and p=0.25, which probability is inside the negative logarithm in binary cross-entropy?
5. A softmax implementation overflows for logits [1000,1001]. Which change preserves probabilities while preventing this overflow?

Select one answer.

Join Discord