Explain the regularization techniques used in deep learning, including dropout, batch normalization, L1/L2 regularization, and data augmentation. How do they prevent overfitting, and when would you use each?

Sample answer preview

Regularization techniques prevent overfitting by constraining the model or augmenting the data to improve generalization. Each technique works differently and is suited to different situations. Dropout randomly sets a fraction of neuron outputs to zero during training.