You have trained a deep neural network model on Google Cloud. The model has low loss on the training data, but is performing worse on the validation data. You want the model to be resilient to…

Question

Options

• AApply a dropout parameter of 0.2, and decrease the learning rate by a factor of 10.
• BApply a L2 regularization parameter of 0.4, and decrease the learning rate by a factor of 10.
• CRun a hyperparameter tuning job on AI Platform to optimize for the L2 regularization and dropout
• DRun a hyperparameter tuning job on AI Platform to optimize for the learning rate, and increase the

• A
  13% (8)
• B
  8% (5)
• C
  77% (49)
• D
  3% (2)

Explanation

When a model performs well on training data but poorly on validation data, it is overfitting. The two primary regularization tools to combat overfitting in deep neural networks are dropout (randomly deactivates neurons during training) and L2 regularization (penalizes large weights). Rather than manually guessing values for these hyperparameters (as Options A and B do with specific fixed values that may not be optimal), running a hyperparameter tuning job on AI Platform systematically searches the hyperparameter space to find the best combination of dropout and L2 regularization values. This is the most principled and efficient approach. Option D is incorrect because tuning only the learning rate and increasing the number of layers would not address overfitting - it could make overfitting worse.

No community discussion yet for this question.