Question 33
Domain 2 — Data, Machine Learning, and Model DevelopmentHow does batch normalization help in training deep neural networks?
Correct answer: B
Explanation
Batch normalization stabilizes layer inputs during training, which reduces internal covariate shift and makes optimization easier. Because the network can use larger learning rates and gradients behave more predictably, it often leads to "speeding up convergence" in deep neural networks.
Why each option is right or wrong
A. By reducing overfitting
B. By speeding up convergence
Batch normalization normalizes activations within a mini-batch to keep their distribution more stable during training, which reduces internal covariate shift and makes gradient-based optimization easier. In practice, this lets the model train with larger learning rates and reach a good solution in fewer iterations, so the training process converges faster.
C. By increasing model accuracy
D. By simplifying the model architecture