Login to unlock
Deep Learning · Phase 2: Forward & Backward Propagation · Loss Functions for Deep Networks

Loss Functions for Deep Networks — Checkpoint Quiz

Checkpoint quiz 3 questions
Back to course

Check your understanding of this module before moving on.

?

Checkpoint quiz 3 questions

A quick check on this module's key ideas before you move on
0/3
Question 01

Why is cross-entropy loss generally preferred over mean squared error for classification problems with a softmax output?

A is correct. When combined with softmax, cross-entropy's gradient simplifies to (predicted - true), which stays informative even for very wrong, confident predictions. MSE through a softmax can saturate and produce tiny gradients in the same scenario, slowing learning.
Question 02

What does a softmax layer do to a vector of raw output scores (logits)?

B is correct. Softmax turns arbitrary real-valued logits into a probability distribution: it exponentiates each value (so everything becomes positive) and divides by the sum (so they total 1).
Question 03

For a regression task predicting a continuous value, which loss function is the standard default and why?

B is correct. MSE's gradient is proportional to (predicted - actual), so bigger mistakes produce proportionally bigger gradient signals, which is exactly the behavior wanted when predicting a continuous quantity.