An ML engineer is training a simple neural network model. The ML engineer tracks the performance of the model over time on a validation dataset. The model's performance improves substantially at first

Question

Options

• AEnable early stopping on the model.
• BIncrease dropout in the layers.
• CIncrease the number of layers.
• DIncrease the number of neurons.
• EInvestigate and reduce the sources of model bias.

• A
  77% (17)
• C
  5% (1)
• D
  5% (1)
• E
  14% (3)

Explanation

This scenario describes overfitting - the model learns the training data so well it loses the ability to generalize, causing validation performance to initially rise then fall.

Why A and B are correct:

• A (Early stopping) directly addresses overfitting by halting training at the epoch where validation performance peaks, before degradation begins - it's the most targeted fix.
• B (Increase dropout) randomly deactivates neurons during training, forcing the network to learn more robust, generalized features rather than memorizing training patterns.

Why the distractors are wrong:

• C & D (More layers / more neurons) increase model capacity, which makes overfitting worse, not better - a larger model has more parameters to memorize noise.
• E (Reduce model bias) addresses underfitting (poor training performance from a model that's too simple), which is the opposite problem from what's described here.

Memory tip: Think of the validation curve as a hill - overfitting is when you go over the peak. Both correct answers are "brakes": early stopping literally stops the car, and dropout slows the engine down by randomly disabling it. Adding layers/neurons is stepping on the gas.

No community discussion yet for this question.