A SAS DLPy image classification model achieves high training accuracy but poor validation accuracy. A practitioner adds Dropout layers after the dense layers. What role do Dropout layers play during training?
Select an answer to reveal the explanation.
Short Explanation
Here's the deal — Dropout layers randomly deactivate a fraction of neurons during each training step, so you stop neurons from co-adapting and force the network to spread knowledge across multiple pathways. That acts like an ensemble of subnetworks and helps reduce overfitting.
Full Explanation
The correct role is that Dropout randomly sets a fraction, often called the drop rate, of neuron outputs to zero during each training step. This prevents neurons from co-adapting, forces the network to build redundant representations and distribute knowledge across multiple pathways, and acts as an ensemble of many subnetworks, which helps reduce overfitting. The learning-rate option is wrong because learning rate scheduling is separate. The zero-mean, unit-variance option describes Batch Normalization. The permanent-removal option is wrong because Dropout is random and temporary during training, not permanent pruning.