Case context: You are developing a speech recognition neural network. The current model exhibits high avoidable bias on the training set. You decide to double the number of hidden layers to fix the bias, but after retraining, you notice that the model now suffers from high variance.

Question: Based on the principles of diagnosing and addressing bias/variance, what is the most appropriate next step to resolve the new variance issue without reverting your model size increase?

Sample answer: The most appropriate next step is to apply regularization techniques. Regularization will usually eliminate the increase in variance caused by the larger model size.

Key points:

• Doubling layers addressed the initial high avoidable bias.
• The size increase caused the model to overfit (high variance).
• Regularization is the recommended solution to eliminate this variance increase.

Rubric: Full credit given for correctly diagnosing the solution to the newly introduced variance as applying regularization.