Cross-Entropy Loss

Easy

neural-networks Google Meta ByteDance

Why is cross-entropy loss preferred over mean squared error (MSE) for classification tasks with a softmax output layer?

Multiple Choice

Cross-entropy is computationally cheaper to evaluate than MSE Cross-entropy produces stronger gradients when the prediction is far from the target, avoiding slow learning from saturated outputs MSE cannot be used with softmax because their gradients are incompatible Cross-entropy guarantees convergence to the global optimum in convex loss landscapes

Correct!

Incorrect — try again next time!

Upgrade to Pro to unlock solutions and explanations

Cross-Entropy Loss

Multiple Choice

Related problems

Deep-dive guides